D-CE64-CYCLE-0 + D-PUZZLE-ATTN-0: cross-cycle state, and entropy as the focus of attention on crosswords

.claude/board/LATEST_STATE.md

@@ -1,3 +1,8 @@
+## 2026-10-07 — lance-graph #1395 MERGED (`5658c10`); D-CE64-CYCLE-0 on branch
+
+- #1395: D-CE64-LOOP-0 and D-CE64-STAUNEN-0, probes in `lance-graph-planner/examples` (no contract change).
+- D-CE64-CYCLE-0 (branch `claude/epistemicstate5-ce64-bits-3oe0dq`, unmerged): the cross-cycle executor experiments, extending `staunen_scheduler_probe.rs`; no contract change.
+
 ## 2026-10-07 — lance-graph #1391 MERGED; D-PEARL-PROD-0 shipped
 
 - #1391 (`53212db`): Pearl's ladder in production (`lance_graph_planner::pearl::{reason, revise, hydrate}`; `revise(&Measured, Reading)` writes back only to the measured edge). Supersedes the "on branch, unmerged" line in the section below.

.claude/board/PR_ARC_INVENTORY.md

@@ -1,3 +1,10 @@
+## 2026-10-07 — lance-graph PR #1395 (merged `5658c10`, branch `claude/epistemicstate5-ce64-bits-3oe0dq`, head `086189a`, 5 commits) — D-CE64-LOOP-0 + D-CE64-STAUNEN-0: self-orchestrating fold loop, surprisal scheduler, bits 40..63 census
+
+- **Added:** `lance-graph-planner/examples/self_orchestration_probe.rs` (one register, five policies, 15 tests) and `staunen_scheduler_probe.rs` (many basins across windows, surprisal distinct from Shannon H, 18 tests), both `test = true`; the bits 40..63 reader/writer census (board entry).
+- **Locked (test-pinned):** naive raw surprise is captured by noise; habituation + learning progress fix it; bits 50..52 are not written (`learn` gates on plasticity); a re-measured channel replaces its evidence. 24 disable runs red.
+- **Deferred / open:** one world per probe; Moore/EWA/Sudoku folds not wired; the `staunen` name collision with `basin_resonance`.
+- **Confidence (2026-10-07):** the single-seed "wake 4.2 vs 5.1" advantage did not survive pooling three seeds (D-CE64-CYCLE-0).
+
 ## 2026-10-07 — lance-graph PR #1393 (merged `69e661a`, branch `codex/ce64-semantic-upper-half-probe`, head `6c98015`, 6 commits) — D-CE64-SEM-0: a probe-local semantic reading of the CE64 upper half
 
 - **Added:** test-only readings in `pearl_ladder_probe.rs`: bits 46..49 as a signed reaction activation, 50..52 as novelty (`ceil(log2)` of the surviving `EpistemicState5` population; `None` when no state survives), 53..58 as a belief-update breadcrumb. Tests ported onto `planner::pearl` after #1391.

.claude/board/STATUS_BOARD.md

@@ -48,8 +48,9 @@ Plan: `.claude/plans/2026-10-06-global-sudoku-replayable-orchestration-v1.md`. O
 | **D-EPI-CANON-0** | Canonical joint `EpistemicState5` reading of bits 59..63 in the contract | Shipped (#1381) | `contract::epistemic_state5`: Cartesian `Topology2 × Certification3`, `raw5 = t \| c << 2`, 24 meaningful / 8 reserved; per-factor facts; declarations + `project_state5`; `CausalEdge64::{epistemic_raw5, with_epistemic_raw5}`; non-declared-reading writers deprecated |
 | **D-EPI-POP-0** | Facts → 32-state population: the missing operator between one edge's coordinate (u5) and a semantic condition over all coordinates (u32) | Shipped (#1381) | `contract::epistemic_state5::{Population, facts_population, EpistemicState5::bit}`: `const fn`, no DSL / query object / planner. Tests pin exact sets (`CAUSES` = {20..23}, `IND_UNKNOWN \| RELATED` = {10,14,18,22}, `IND_KNOWN \| SUPPORTS` = {13,17,21}, `RELATED` = 8..23), operator == state enumeration over all 2048 requirements with no reserved code, conjunction == intersection; 2 disable runs red (any-of semantics; reserved read as fact-less) |
 | **D-EPI-LEGACY-DEPROJECT-0** | Deproject the remaining `ReasoningBand` / CE64-`TrustTexture` consumers: each goes to an exact coordinate, a `facts_population(..)`, MUL, or another subsystem (e.g. counterfactual context) — never a 1:1 replacement enum | Shipped (#1381) | `entries/2026-10-07-epi-legacy-deprojection.md`: 6 ReasoningBand caller sites (1 certification question, 1 non-interference witness → both exact coordinates; 2 context/obsolete → compat; 2 refusal witnesses), 0 production CE64 `TrustTexture` consumers, 4 `TrustTexture` types with no bridge; `EdgeRole::band` removed; hinge + V3 lift migrated; 4 code loops → `facts_population` |
-| **D-CE64-STAUNEN-0** | Surprisal scheduler over many CE64 basins across windows (`lance-graph-planner/examples/staunen_scheduler_probe.rs`) | In PR | naive raw surprise is captured by noise (resolved 0.649); habituation + learning progress fix it (0.952, wake 4.2 windows vs round-robin 5.1); round-robin still resolves slightly more (0.964); bits 50..52 left as plasticity (`learn` gates on them); 18 tests, 13 disable runs red (`entries/2026-10-07-ce64-staunen-0-surprisal-scheduler.md`) |
-| **D-CE64-LOOP-0** | Bounded self-orchestrating fold loop over one CE64 register (`lance-graph-planner/examples/self_orchestration_probe.rs`) + bits 40..63 census | In PR | combined policy reaches `IndirectKnown × Causes` with H ≤ 0.5 bit in 3 folds (round-robin 10, random median 14); ~830 ns/fold release; 15 tests in CI, 11 disable runs red; H and activation stay out of the register (`entries/2026-10-07-ce64-loop-0-self-orchestration.md`) |
+| **D-CE64-CYCLE-0** | Cross-cycle executor experiments on the surprisal scheduler: forget arms, FIFO baseline, 350 ms windows, 1k..1M folds, attractors | In PR | the next cycle needs evidence table + register (Epi5) + attention state; H and bits 40..42 are derived/echo; pooled 3 seeds: no wake advantage over round-robin, 11% fewer folds; FIFO fits 1M folds in a 350 ms window at ~220 ns/fold, the heap scheduler ~440k at ~800 ns (`entries/2026-10-07-ce64-cycle-0-cross-cycle-executor.md`) |
+| **D-CE64-STAUNEN-0** | Surprisal scheduler over many CE64 basins across windows (`lance-graph-planner/examples/staunen_scheduler_probe.rs`) | Shipped (#1395); wake claim corrected by D-CE64-CYCLE-0 (pooled: no wake advantage) | naive raw surprise is captured by noise (resolved 0.649); habituation + learning progress fix it (0.952, wake 4.2 windows vs round-robin 5.1); round-robin still resolves slightly more (0.964); bits 50..52 left as plasticity (`learn` gates on them); 18 tests, 13 disable runs red (`entries/2026-10-07-ce64-staunen-0-surprisal-scheduler.md`) |
+| **D-CE64-LOOP-0** | Bounded self-orchestrating fold loop over one CE64 register (`lance-graph-planner/examples/self_orchestration_probe.rs`) + bits 40..63 census | Shipped (#1395) | combined policy reaches `IndirectKnown × Causes` with H ≤ 0.5 bit in 3 folds (round-robin 10, random median 14); ~830 ns/fold release; 15 tests in CI, 11 disable runs red; H and activation stay out of the register (`entries/2026-10-07-ce64-loop-0-self-orchestration.md`) |
 | **D-PEARL-PROD-0** | Pearl's ladder in production: `lance_graph_planner::pearl::{reason, revise, hydrate}` over `lance_graph_contract::certification` | Shipped (#1391) | the P7a folds become `contract::certification::{CertificationModel, ModelBuilder}`; the dispatch, write-back and topology gate become `planner::pearl`, taking the caller's declared reading. The 14 falsifier tests run in the planner suite (CI); 12 disable runs red against the production module. No runtime caller yet: nothing in production produces sealed populations or chains for it (`entries/2026-10-07-pearl-prod-0-production-wiring.md`) |
 | **D-PEARL-IO-0** | Pearl's ladder executed in and out of `CausalEdge64`: bits 40..42 select the operator, the measurement revises bits 59..63 | Shipped (#1390) | `cognitive-shader-driver/examples/pearl_ladder_probe.rs`, `entries/2026-10-07-pearl-io-0-ladder-in-and-out-of-ce64.md`: SO → P7a observational folds, PO → P7a executed arms (only path to `Causes`), SPO → `counterfactual_replay` (reaction, never a certification), SP → SO-vs-PO direction; `hydrate` gates `IndirectUnknown → IndirectKnown` on both bindings. CUI BONO script: `IndirectUnknown × Related` → `IndirectKnown × Related` → `IndirectKnown × Causes`. 14 tests, 12 disable runs red. P7a model moved to `examples/shared/certification_model.rs` unchanged |
 | **D-PUZZLE-0** | Sudoku / crossword / chess on the same population algebra under domain laws | In progress | step 1 Sudoku shipped (#1382). Step 2 crossword (`crossword_population_fold_probe.rs`, `entries/2026-10-07-puzzle-0-crossword-same-algebra.md`): domain-free half moved to `examples/shared/population_fold.rs` (fence test, Sudoku counts unchanged); 1,000,052 crossword edges, counts equal 3 ways; filter 0.30–0.43 ns/edge in both domains. Step 2b real words (`crossword_real_words_probe.rs`, `entries/2026-10-07-puzzle-0-crossword-real-words.md`): English = DeepNSM v1 4096 + academic_20k (committed), German = DeReKo top 20k only from `DEREKO_PATH` (CC BY-NC, nothing committed); ~1.0M edges each, counts equal 3 ways. Steps 2/2b merged (#1384). Step 3, merged #1387 (`crossword_mask_propagation_probe.rs`, `entries/2026-10-07-puzzle-0-crossword-mask-propagation.md`): Morton cell + `(slot:offset)` lanes, DeepNSM-v2 `WordId` masks, `MooreSymbol8` letters; four arms on identical puzzles (EN 5×5: token 3.0 µs ≈ hybrid 2.9, Cartesian 5.1, literal ~1,045); with no repeated words, creation stops at 5×5 with the 18–20k vocabularies; cui-bono ablation of givens. Next: chess in `stockfish-rs` (GPL) — how it reaches the shared half is open |

.claude/board/entries/2026-10-07-ce64-cycle-0-cross-cycle-executor.md

@@ -0,0 +1,78 @@
+# 2026-10-07 — D-CE64-CYCLE-0: what a CE64 register must carry across cycles, and what self-orchestration buys at scale
+
+**Status:** TEST-PINNED (`crates/lance-graph-planner/examples/staunen_scheduler_probe.rs`, `test = true`, 23 tests run by `member-tests`; comparative tests pool 3 seeds). MEASURED (`cargo run --release -p lance-graph-planner --example staunen_scheduler_probe`, 4-core Xeon 2.8 GHz). No production code changed.
+
+## DECISION
+
+- **SCOPE:** the D-CE64-STAUNEN-0 executor extended rather than duplicated: `run_with(opts)`, a lazy-stamp heap scheduler, a FIFO worklist baseline (reacts to new evidence, reads no entropy or activation), a wall-clock budget per window with continuation, `Forget` arms at window boundaries, the #1390 Simpson population plus its positive trial, end-attractor and fold-quadrant readings.
+- **The register is treated as the cross-cycle carrier; which state it must carry is measured by forgetting each piece at every window boundary** (single seed, 2000 windows, budget 3):
+
+| forgotten | resolved | same run? | verdict |
+|---|---|---|---|
+| nothing | 0.955 | — | — |
+| posterior (H) | 0.955 | identical | derived from the evidence table: persisting H would be an echo |
+| bits 40..42 (question) | 0.955 | identical | an echo of the opcode in this loop |
+| register (Epi5, mask) + what ran at which code | 0.024 | no | earned Epi5 is re-derived every window and starves the real work |
+| evidence table (latest outcome per channel) | 0.000 | no | needed |
+| attention state (habit, progress, coherence, surprise) | 0.378 | no | needed; this is how earlier windows direct later ones |
+
+- **W:** belief moved on 4648 folds while Epi5 moved on 72; `pearl::revise` changed F/C on 0. A W that advances only on Epi5 changes would reference about 1.5% of belief updates. Nothing here mints a W.
+- **BASIS:** all weights, likelihoods and thresholds are policy pins.
+
+## Measured: self-orchestration against baselines (pooled 3 seeds, 600 windows)
+
+| policy | folds | resolved | wake | folds on settled basins | resolved basin-windows per fold |
+|---|---|---|---|---|---|
+| full | 4806 | 0.865 | 7.9 | 0.047 | 10.36 |
+| FIFO worklist | 5400 | 0.909 | 8.3 | 0.094 | 9.69 |
+| round-robin | 5400 | **0.911** | **7.5** | 0.093 | 9.72 |
+| random | 5400 | 0.838 | 12.4 | 0.050 | 8.94 |
+| no entropy and no activation | 5400 | 0.486 | 11.4 | 0.018 | 5.18 |
+
+- **Correction to D-CE64-STAUNEN-0 / #1395:** its "wake 4.2 vs round-robin 5.1" was one seed. Pooled, the full scheduler has **no wake advantage** (7.9 vs 7.5). Exact tie-breaking alone had moved a single-seed wake latency by half.
+- What survives pooling:
+  - ablations each matter (no entropy 0.515, no surprise wake 43.1, no habituation 0.712, no progress 0.822, naive 0.629);
+  - activation drives sleeping (without it, 5400 folds instead of 4806);
+  - the full scheduler spends 11% fewer folds and wastes half as many on settled basins, at a lower resolved fraction than the two simple baselines.
+
+## Measured: dense windows, 350 ms cap
+
+| folds per window | policy | ns/fold | folds that fit in 350 ms | productive share |
+|---|---|---|---|---|
+| 1M (327,680 basins) | FIFO / round-robin | ~220 | all 1M (~252 ms per window) | 0.72 |
+| 1M (327,680 basins) | full (heap) | ~800 | ~440k | 0.89 |
+| 1k–100k | FIFO / round-robin | ~210 | — | ~1.00 |
+| 1k–100k | full (heap) | 490–710 | — | 0.88 |
+
+- At 1M, FIFO yields about twice as many productive folds per second as the heap-driven scheduler.
+- **Within a window, later folds are less productive** (FIFO: first tenth 0.953, last tenth 0.483). Informative work is front-loaded and runs out.
+- **"Does fold 1,000,000 get better directed by folds 1..999,999?"** Inside a window, no. Across windows, the carried attention state is what directs later windows (forgetting it: 0.955 → 0.378 resolved).
+
+## Other readings
+
+- Fold quadrants (activation sign × ΔH, full): falsify 1794, anomaly 1935, leave 419, dig 233, other 758.
+- Polarity: hydration reacts negative / zero / positive (2233 / 561 / 2168). The SP check is negative only on Simpson basins (pinned with a silence twin).
+- End attractors (learned, wrong-settled, exhausted, waiting, sterile, active):
+  - full: [31, 0, 0, 0, 7, 2]
+  - naive: [13, 13, 0, 4, 7, 3]. Naive scheduling ends **wrongly settled** as often as learned: an attractor is not understanding, and the reading distinguishes them.
+- **The sterile class is weak:** on the 600-window shape only 3 of 8 noise basins meet it.
+
+## Gates
+
+19 disable runs, each red on its named falsifier: 13 from D-CE64-STAUNEN-0, re-run after the refactor, and 6 new:
+- the forgotten belief not rebuilt
+- forgetting the register as a no-op
+- the W counter dropped
+- Simpson replaced by a plain trial
+- a deadline that never binds
+- no sterile class
+
+## OPEN
+
+- Within-window direction does not improve; front-loading is the dominant dynamic.
+- The heap scheduler costs ~3.5× FIFO per fold. A cheaper priority structure (bucketed priorities, Moore-local queues) is untested.
+- Moore, EWA, Sudoku and Quack folds are not in the loop (they live outside the planner). Substrate-agnosticism is therefore claimed for nothing beyond Pearl folds here.
+- The register itself carries none of the needed cross-cycle state except Epi5:
+  - The evidence table, which drives this loop, lives beside the register; it is 4 channels × 2 bits.
+  - Whether that state, attention, and a reference to belief updates should have a home in the register remains open; bits 50..52 conflict with `learn` (D-CE64-STAUNEN-0).
+- No counterfactual (−6) fold in this loop; isolation is pinned in D-CE64-LOOP-0 and the planner's Pearl tests.

.claude/board/entries/README.md

@@ -25,7 +25,7 @@ index row, (3) no duplicate entry id. Checks 1 and 2 are deliberately
 opposite directions; the stranding this convention prevents shows up in
 exactly one of them, never both.
 
-259 entries, 2026-08-06 .. 2026-10-07.
+260 entries, 2026-08-06 .. 2026-10-07.
 
 | date | entry id | finding | file |
 |---|---|---|---|
@@ -49,6 +49,7 @@ exactly one of them, never both.
 | 2026-10-07 | `D-CTX-7` |  | [2026-10-07-ctx7-revision-writer-for-epistemic-state.md](2026-10-07-ctx7-revision-writer-for-epistemic-state.md) |
 | 2026-10-07 | `D-CE64-STAUNEN-0` |  | [2026-10-07-ce64-staunen-0-surprisal-scheduler.md](2026-10-07-ce64-staunen-0-surprisal-scheduler.md) |
 | 2026-10-07 | `D-CE64-LOOP-0` |  | [2026-10-07-ce64-loop-0-self-orchestration.md](2026-10-07-ce64-loop-0-self-orchestration.md) |
+| 2026-10-07 | `D-CE64-CYCLE-0` |  | [2026-10-07-ce64-cycle-0-cross-cycle-executor.md](2026-10-07-ce64-cycle-0-cross-cycle-executor.md) |
 | 2026-10-07 | `bind-bundle-gated-gather` | One bound query, five routes, identical answers; the semijoin is never gated and dominates; a gated mask gather beats a selection vector (R1 survives) | [2026-10-07-bind-bundle-gated-gather.md](2026-10-07-bind-bundle-gated-gather.md) |
 | 2026-10-07 | `aperture16-u64-word-schedule` | The u16 view of the 64K mask is free but 1.3–2.3× slower per layout as a schedule than u64 words; extent × mask compose; the exact next-frontier mask beats a u16 target histogram; no new carrier or opcode | [2026-10-07-aperture16-u64-word-schedule.md](2026-10-07-aperture16-u64-word-schedule.md) |
 | 2026-10-06 | `v3-v4-dual-reading-round3` |  | [2026-10-06-v3-v4-dual-reading-round3.md](2026-10-06-v3-v4-dual-reading-round3.md) |