diff --git a/.claude/board/LATEST_STATE.md b/.claude/board/LATEST_STATE.md index 38d5f6665..9c197c9db 100644 --- a/.claude/board/LATEST_STATE.md +++ b/.claude/board/LATEST_STATE.md @@ -1,3 +1,12 @@ +## 2026-10-26 — main after #1792 (merged `30ca36ec0`) + +- D-DQM: Lance 13 crosswalk and pre-registered probes (#1788); P1 chunk-touched byte predictor FAILED (#1789). +- D-DR256 labs in `drf-focus-probe`: one global 8 KiB mask from 64 mapped registers (#1790); on-demand rebuild vs bounded reuse, u16 scatter, L1 interference (#1791, with ndarray #377 `mask_scatter_or_u16`). +- D-DR256-WHERE capstone (#1792): WHERE/COUNT/ANY already exist in Quack/mask-risc; reused R costs 0.05–0.60 of recompute; stale inputs run silently stale. D-NFR-1: in-borrow execution needs no frame; escaped reuse must recompute or refuse. +- Open: NFR-OP-1 (escaped-reuse authority contract), S-INGRESS, one `_rowid` authority, mask-risc `LaneRef::U16`, P3–P5. +- **Contract inventory delta:** none (lab examples, test-only falsifiers, docs). +- Per-PR records: `PR_ARC_INVENTORY.md` catch-up #1788–#1792. Not recorded: #1772–#1787. + ## 2026-10-25 — main after #1771 (merged `332b420c`) - New lab crate on main: `drf-focus-probe` (#1762, workspace-excluded, no CI job). It carries the D-DRF-1 focus lease + gather (DRF-P1), the focus-vs-direct break-even (DRF-P2, #1763/#1764) and the writeback falsifiers W1–W4, W6–W8 (#1765–#1771). Feature `handoff` hands flushed rows to the real `lance-graph-supervisor` CycleWriter; a fake WAL stands in for Lance. diff --git a/.claude/board/PR_ARC_INVENTORY.md b/.claude/board/PR_ARC_INVENTORY.md index d12309f27..82ee6db75 100644 --- a/.claude/board/PR_ARC_INVENTORY.md +++ b/.claude/board/PR_ARC_INVENTORY.md @@ -1,3 +1,41 @@ + + +## 2026-10-26 — lance-graph PR #1792 (merged `30ca36ec0`, branch `claude/dr256-where-capstone`, head `2971b870`) — D-DR256-WHERE: capstone, DR64³ inventory, P2 Quack-path lab, evidence reconciliation, D-NFR-1 ownership correction + +- **Added:** plan `dr256-where-capstone-v1.md` (P0 census, Amendment A DR64³ inventory, B P2, C compiler compatibility, D reconciliation, E D-NFR-1); `drf-focus-probe/examples/{where_occupancy,p2_quack_where}.rs`; test-only falsifiers `soa-view-probe/tests/falsifiers.rs` (escaped mask vs packet cycle), `cognitive-shader-driver/tests/escaped_reuse.rs` (6), `lance-graph-planner/tests/horizon_is_not_a_snapshot.rs` (2). +- **Locked:** in-borrow Quack execution needs no frame store; the current cycle/write stamps cannot validate an escaped `Keep` mask; `QueryReference::at` is a horizon, not a physical snapshot. +- **Deferred:** escaped-reuse authority contract (NFR-OP-1, narrowed); production binding store; P3–P5; ScatterOr unchanged. +- **Docs:** entry `2026-10-26-dr256-where-capstone.md`; `d-native-repr-contract-v1.md` NFR-OP-1 row. +- **Confidence:** lab; each falsifier disable-verified, one non-binding disable recorded; no production code. + +## 2026-10-26 — lance-graph PR #1791 (merged `cd4901122`, branch `claude/dr256-on-demand-lab`, head `2cf36939`) — lab(D-DR256-OD): on-demand global mask, u16 scatter, bounded reuse, L1 interference + +- **Added:** `drf-focus-probe/examples/dr256_on_demand.rs` (arms A–E, N ∈ {1,2,4,8,64}); depends on ndarray #377 `mask_scatter_or_u16`. +- **Deferred:** mask-risc `LaneRef::U16` (designed, not built); a witnessed derived-population contract for reusing a scattered G (proposal only). +- **Docs:** entry `2026-10-26-dr256-on-demand-l1.md` (pre-registered `e01d38879`). +- **Confidence:** lab; rebuild-per-thought reaches ms at chaotic/full N = 64 (a failed gate); timings reported, not gated. + +## 2026-10-26 — lance-graph PR #1790 (merged `3830229dd`, branch `claude/dr256-global-mask-lab`, head `a68bf8e7`) — lab(D-DR256): 64 mapped DR64 registers -> one global 8 KiB mask + +- **Added:** `drf-focus-probe/examples/global_union.rs` (oracle, ndarray ×64, one mask-risc program, `G AND h`); dev-dependency `lance-graph-mask-risc`; index-XOR and gate-drop negative controls asserted on every run. +- **Locked:** further Boolean algebra over a scattered G is blocked by the `ScatterOrU32` survival condition (`ir.rs:329-337`, unchanged). +- **Docs:** entry `2026-10-26-dr256-global-union-lab.md`. +- **Confidence:** lab; controls made permanent after Codex review. + +## 2026-10-26 — lance-graph PR #1789 (merged `ab1f3a83f`, branch `claude/dqm-p1-chunk-touched`, head `08a8bcb6`) — probe(D-DQM P1): chunk-touched predictor vs H14c/H14d Lance bytes — FAIL + +- **Added:** `.claude/probes/p1-chunk-touched-v1/{p1.py,results.csv}` (offline, committed data only). +- **Locked:** verdict FAIL on the registered can-fire control (clustered 1 %: predicted 0.44, measured 0.24 / 0.19); no estimator proposed. +- **Docs:** entry `2026-10-26-dqm-p1-chunk-touched.md`, pinned to registration `62523c0e4`. +- **Confidence:** measured; registered kill thresholds pass, the FAIL rests on the control (stated after Codex review). + +## 2026-10-26 — lance-graph PR #1788 (merged `10291a8cc`, branch `claude/dqm-lance13-crosswalk`, head `c33eb056`) — docs(D-DQM): DR64 ↔ DuckDB ↔ Lance 13 capability crosswalk + +- **Added:** Appendix A in `duckdb-dr-quack-machine-room-reuse-v1.md` (A.0–A.10, probes P1–P5 pre-registered in A.6). +- **Deferred:** S-INGRESS; one coordinate → `_rowid` authority; P2–P5. +- **Docs:** entry `2026-10-26-coresearch-lance13-masked-read-dr64.md`; `INTEGRATION_PLANS.md`. +- **Confidence:** docs; unread claims marked UNVERIFIED. + ## 2026-10-25 — lance-graph PR #1771 (merged `332b420c`, branch `lab/drf-w6-resample-isolate-20261025`, head `7125af32`) — test(drf-w4/w6): isolate the resampling refusal to changed bytes