From 79d4b10f234617ec158f574b1fcc6030e0181480 Mon Sep 17 00:00:00 2001 From: Claude Date: Thu, 15 Oct 2026 21:32:50 +0000 Subject: [PATCH] board: #1566 and #1567 merged (arc entries, latest state) Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh --- .claude/board/LATEST_STATE.md | 5 +++++ .claude/board/PR_ARC_INVENTORY.md | 14 ++++++++++++++ 2 files changed, 19 insertions(+) diff --git a/.claude/board/LATEST_STATE.md b/.claude/board/LATEST_STATE.md index b5d09bad5..fd3d746fb 100644 --- a/.claude/board/LATEST_STATE.md +++ b/.claude/board/LATEST_STATE.md @@ -1,3 +1,8 @@ +## 2026-10-15 — main after #1567 (merged `b1f1365d`); D-HOLO H6c/H14 and the writer's version clock on main + +- **On `main` (#1566, merged `c18aefcb`):** D-HOLO H6c and H14 probes and results; H14b open (awaits the operator). +- **On `main` (#1567):** `LanceCycleWriter::version_clock()` follows the #1562 contract (seeded at the head on open, positions not live rows). Reader not wired. + ## 2026-10-15 — main after #1565 (merged `2379ddc2`); D-HL-5 probes on main - **On `main` (#1564, merged `703d6a94`):** six D-HL-5 thought-lane probes in `cognitive-shader-driver/examples` and the givens layer (hydration writes thought 0; contradicting hypotheses refused and recorded). Probes only; no production wiring. diff --git a/.claude/board/PR_ARC_INVENTORY.md b/.claude/board/PR_ARC_INVENTORY.md index a84bcbd06..cd102ce93 100644 --- a/.claude/board/PR_ARC_INVENTORY.md +++ b/.claude/board/PR_ARC_INVENTORY.md @@ -1,3 +1,17 @@ +## 2026-10-15 — lance-graph PR #1567 (merged `b1f1365d`, branch `ccr-9f3b8075-74u47l-version-clock`, head `6ced7887`, 3 commits) — D-WAL-6: the version clock wired into `LanceCycleWriter` + +- **Added:** `LanceCycleWriter::version_clock()`; `sync_clock` after open, bootstrap and every commit: records the new head's row positions (sum of `physical_rows` over fragments), keeps the clock on a no-op, reseeds when history went back. Seven tests in `cycle_sink.rs`, written first (red against a stub returning `None`). +- **Locked:** the #1562 contract: a reopened writer seeds the clock at its head and answers `None` for older versions; deleted rows keep their positions, so the clock counts positions, not live rows; a reconciled append (published, then failed) still records its version. +- **Deferred / open:** the reader is not wired; a rewrite that reorders rows needs a new clock. +- **Confidence:** high; three disable runs red (no sync after commit, live rows instead of positions, reseed on every sync). + +## 2026-10-15 — lance-graph PR #1566 (merged `c18aefcb`, branch `ccr-9f3b8075-74u47l-holo-h6c`, head `09cf22c0`, 13 commits) — D-HOLO H6c measured, H14 lane storage measured + +- **Added:** `holo_tile_crossover_probe` (H6c) and `holo_lane_storage_probe` (H14) in `lance-graph-mask-risc/examples`; plan sections and results for both; entries `2026-10-14-d-holo-h6c.md`, `2026-10-15-d-holo-h14-lane-storage.md`; STATUS rows D-HOLO-H6c, D-HOLO-H14. Board catch-up for #1564/#1565. +- **Locked:** H6c (v4, three runs): Q1 holds, P/D crosses 1 between g = 8 and 12 (once 12 → 16) in both layouts; Q2 no switch certified (G = 8 worst query 1.21–2.28× D), the byte switch unsafe (9.8–15.1×); Q3 declared build passes (better of P/H 0.32–0.42× D). H14 (v4): C's folds beat B by 10–24× (F1) and 39–256× (F2); C's merge-on-read full-row read is 2.8–6.1× slower than B; neither C nor B is the default by the pre-registered rule. +- **Deferred / open:** H14b (C with a per-row lane mask so the merge reads only the touched lanes) awaits the operator; C's seal at low dirt (page faults); H14 timed on one class only. +- **Confidence:** medium; oracle tests and disable runs red; about 40 % run-to-run noise on the VM. Codex P2s fixed in review: the crossover is now an adjacent below-to-above step; C's F2 grows with thoughts. + ## 2026-10-15 — lance-graph PR #1565 (merged `2379ddc2`, branch `ccr-9f3b8075-74u47l-r1-test`, head `39ad8b1a`, 1 commit) — contract: the R1 test checks every verdict from `CognitiveWork` - **Added:** `kanban.rs` R1 test walks all nine (effect, counterfactual) verdicts from `CognitiveWork`: none reopens the goal, the one accepted verdict moves on to `Evaluation`, the other eight leave the column unchanged.