Levius-Fubuki · Oct 28, 2026 · Oct 28, 2026 · Oct 28, 2026 · Oct 28, 2026 · Oct 28, 2026
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -49,6 +49,10 @@ jobs:
         run: |
           python -m pip install pytest
           PYTHONPATH=src python -m pytest tests/clef -q
+      - name: D3 legacy and turbo reference profile tests
+        run: |
+          python -m pip install pytest pillow
+          PYTHONPATH=src python -m pytest tests/d3 -q
 
   cuda-contract:
     # The CUDA backend contract and its checker. No CUDA toolkit and no GPU: both

diff --git a/README.md b/README.md
@@ -192,6 +192,7 @@ checkpoint. Public231 concurrency-1 latency for heads `1ba07d1`, `f968e576`,
 | LFM2.5-350M | [Transformers worker](recipe/lfm2/README.md); text state and independent choice questions |
 | OmniJev-4B v1.1 | [Native Rust/CUDA worker](recipe/omnijev/native.md), eager: one image, Choice, Noul and Score; [reference comparison on sm_89](recipe/omnijev/native.md#comparison-with-the-reference) |
 | D3 (Decision 3.0) | [Python reference worker](recipe/d3/reference.md) on the checkpoint's pinned runtime; `choice`/`noul`/`score` over text, image and video state; author-run on A100 for text and media serving (see the integration and media PRs) |
+| D3 turbo (Decision 3.0) | [Explicit Python reference profile](recipe/d3/turbo.md) for pinned `vllm-sr/d3-turbo`; text/JSON and one PNG/JPEG, `choice`/`noul`/`score`; reference fixtures and Public231 validated on A800 / SM80, native and optimized execution unimplemented |
 | Vela 2.0 0.3B | [Python reference worker](recipe/vela/reference.md) on the checkpoint's pinned runtime; all five kinds (`choice`/`noul`/`score`/`span`/`set`) for text state; CPU FP32 validated (author-run, evidence in the integration PR) |
 | Vela 2.0 0.8B hybrid | [Python reference worker](recipe/vela/reference.md) on the checkpoint's pinned runtime; same surface plus `span_heads`; CPU FP32 parity recorded in the integration PR |
 | Clef-flash (9B) and Clef (27B) | [Rust `omni-clef` server](recipe/clef/reference.md) plus the pinned Python joint-schema executor on one persistent thread; text `choice`/`noul`/`score`; flash is the bring-up variant. Image, video, and a native joint head are not served. The L20X-labeled lab Public231 C1 primary is the previous Python HTTP worker, tip `594528a9`: p50 164.969 ms, p95 216.755 ms, `/health` `fla+causal_conv1d`, versus `9030065` run1 p50 247.518 ms, p95 413.025 ms (Δp50 −82.549 ms, Δp95 −196.270 ms). Operator-reported `omni-clef` tip `f9127a0` on that host is a serve-path paste, not a latency KEEP: R1 p50 169.605 ms, p95 220.003 ms (231/231) does not pass the dual gate versus 164.969/216.755; R2 53.580/97.031 and R3 52.950/97.608 are not a new primary. Smoke 4/4, `/health` `fla+causal_conv1d`, quality matches the baseline (`choice_acc=0.8417`, `noul_acc=0.7973`, `score_mae=0.2488`) |

diff --git a/docs/supported-models.md b/docs/supported-models.md
@@ -28,6 +28,7 @@ Models that are being added are also tracked in issues labeled [new model](https
 | CLM-v0.1-8B | [Native Rust engine](../src/models/clm/README.md), `clm-run`, in front of a Qwen3-8B `/v1/embeddings` server | Engine-only agreement **4.5e-06** on identical vectors ([#29](https://github.com/ThinkFlowLab/system1-omni/pull/29), [validation](../recipe/clm/native/VALIDATION.md)); contract checks documented ([#23](https://github.com/ThinkFlowLab/system1-omni/pull/23)) | **Validated on compute capability 8.9** ([#29](https://github.com/ThinkFlowLab/system1-omni/pull/29), [validation](../recipe/clm/native/VALIDATION.md)) | Unverified | Rust to build and a separate Qwen3-8B `/v1/embeddings` server for the encoder; the engine itself has no device-specific code |
 | OmniJev-4B v1.1 | [Native Rust/CUDA worker](../recipe/omnijev/native.md) on the shared Qwen3.5 vision and language path, eager | Contract, preparation, pixels, heads and finishing checked against the reference; no CPU inference | Compared with the reference on compute capability 8.9 for [140 answers](../recipe/omnijev/native.md#comparison-with-the-reference) | Not supported | Compute capability 8.0 or newer, CUDA toolkit to build, the export from `export.py` |
 | D3 (Decision 3.0) | [Python reference worker](../recipe/d3/reference.md) on the checkpoint's pinned runtime — text, image and video state over `choice`/`noul`/`score`, with the deliberate transport differences documented in the recipe; the native worker is in progress ([#136](https://github.com/ThinkFlowLab/system1-omni/issues/136)) | Unverified | Author-run on NVIDIA A100 (sm_80): serving smoke and reference parity for text and media (exact decisions, max delta 0.0); evidence in the integration and media PRs. Owner: @MrDongsls. | Unverified | Python 3.12, the [pinned runtime requirements](../recipe/d3/requirements.txt), the [pinned package](https://huggingface.co/vllm-sr/d3) (~52 GB); [media budgets and the 64 MiB total request cap](../recipe/d3/reference.md) |
+| D3 turbo | [Explicit Python reference profile](../recipe/d3/turbo.md); fixed public AutoModel, text/JSON and zero or one PNG/JPEG; `choice`/`noul`/`score`. Native and optimized execution are separate milestones | CPU contract/fixture tests; no CPU inference qualification | NVIDIA A800 80GB PCIe / capability 8.0 reference: separate state-once/full-row fixtures, 693/693 Public231 serving and exact official semantic responses at `821d6f933`; pooled p50/p95 233.83/930.21 ms. [Evidence and controls](../recipe/d3/turbo.md#measured-reference-baseline). No native CUDA qualification | Unverified | Fixed package (~52.2 GB), Python 3.12, pinned runtime requirements and 80 GB-class GPU; H200 / SM90 target unverified |
 | Vela 2.0 0.3B | [Python reference worker](../recipe/vela/reference.md) running the checkpoint's pinned runtime | **CPU FP32 validated** (author-run, x86 WSL2, 24 cores): startup with all-kind warmup, the four example requests plus five synthetic rejections, and exact-match reference parity against the pinned upstream server on all nine (four 200s byte-equal incl. key order, agreed 422/413); byte-identical frontend forwarding; evidence in the integration PR. Owner: @MrDongsls. | Unverified (bf16-autocast is an untested deployment option) | Not supported | Python 3.12, the [pinned requirements](../recipe/vela/requirements.txt), the [pinned package](https://huggingface.co/vllm-sr/Vela-2.0-0.3B) (1.24 GB FP32 weights; ~4.4 GB with ONNX/assets) |
 | Vela 2.0 0.8B hybrid | [Python reference worker](../recipe/vela/reference.md) running the checkpoint's pinned runtime (same wire layer as 0.3B) | Same surface as 0.3B plus the `span_heads` field (router/broad dispatch); **CPU FP32 parity** (author-run; evidence in the integration PR) | Unverified (bf16-autocast is the evaluated upstream setting but untested here) | Not supported | Python 3.12, the [pinned requirements](../recipe/vela/requirements.txt), the [pinned package](https://huggingface.co/vllm-sr/Vela-2.0-0.8B) (2.02 GB shard + 4.5 MB broad head; ~3 GB FP32 parameters) |
 | Clef-flash (9B) | [Rust `omni-clef` server](../recipe/clef/reference.md) with the pinned Python executor on one persistent thread; text `choice`/`noul`/`score`; non-thinking prompt; `thinking.enabled: false` admitted and omitted from the prompt; 16384-token refusal. The joint head is still the checkpoint module | Contract tests only, including the persistent-thread serve path; no inference | Kept Public231 primary on the L20X-labeled lab host is the previous `ThreadingHTTPServer` worker. Baseline `9030065` run1 p50 247.518 ms, p95 413.025 ms, 231/231. Tip `594528a9` remeasure, warmup 5, concurrency 1, stock Public231 with `thinking.enabled` false, metric `successful_latency`: 231/231, p50 164.969 ms, p95 216.755 ms, `/health` `fla+causal_conv1d`. Versus run1: Δp50 −82.549 ms (−33.35%), Δp95 −196.270 ms (−47.52%). Operator-reported `omni-clef` tip `f9127a0`: smoke 4/4, `/health` `fla+causal_conv1d`, R1 p50 169.605 ms, p95 220.003 ms (231/231), R2 53.580/97.031, R3 52.950/97.608, quality `choice_acc=0.8417` `noul_acc=0.7973` `score_mae=0.2488` (matches `594528a9`). Not a latency KEEP. R1 does not pass the dual gate versus 164.969/216.755. R2 and R3 are not a new primary. Not re-run here | Not supported | Python 3.12, [card-tested torch 2.11 and transformers 5.10.2](../recipe/clef/requirements.txt), a Rust build of `omni-clef`, pinned `Cloudflare/clef-flash` snapshot (~19.1 GB) |

diff --git a/mkdocs.yml b/mkdocs.yml
@@ -85,6 +85,7 @@ nav:
       - CLM stub-encoder contract: recipe/clm/README.md
       - LFM2.5-350M choice worker: recipe/lfm2/README.md
       - D3 reference worker: recipe/d3/reference.md
+      - D3 turbo reference profile: recipe/d3/turbo.md
       - Vela 2.0 reference workers: recipe/vela/reference.md
       - Clef reference worker: recipe/clef/reference.md
       - pplx-decider reference worker: recipe/pplx_decider/README.md

diff --git a/recipe/README.md b/recipe/README.md
@@ -52,6 +52,7 @@ For a first real decision, follow the [complete CPU walkthrough](../docs/getting
   verify its runtime manifest, and serve `choice`/`noul`/`score` decisions over
   text, image and video state through the Python worker and the Rust frontend
   (text serving and parity validated on A100; media budgets and caps in the recipe).
+- [D3 turbo reference profile](d3/turbo.md): pinned `vllm-sr/d3-turbo` through the existing D3 worker, with text/JSON or one PNG/JPEG, official state-once and typed errors; separate from the legacy profile and native execution.
 - [Vela 2.0 reference workers (0.3B and the 0.8B hybrid)](vela/reference.md): download the pinned
   vllm-sr/Vela-2.0-0.3B package, pass its full-hash package gate, and serve all
   five SystemOne question kinds (`choice`/`noul`/`score` plus the `span`/`set`