From 8b6693213761d662440ae435f910d07b94cbeed1 Mon Sep 17 00:00:00 2001 From: Claude Date: Fri, 16 Oct 2026 04:43:44 +0000 Subject: [PATCH] Ledger revision after #1576; arc entries for #1560, #1570, #1571, #1574-#1576 Maps #1560-#1576 onto the 2026-10-15 ledger correction: nine closed ends (O22 verdict, R1 grid, 59..60 one field, one operand dictionary per epoch, mapping inventory, stamp as evidence lane, version clock, rejected hydrate guard, board catch-up), merged items (O25+K2, N1+decision 2 = D-HL-5 H4, ablation+N15+D-ASM, DOC-WIRE) and new ones (O29, O30, BAND-RETIRE, N16-N21). Adds the six missing PR arc entries and a LATEST_STATE catch-up. Claude-Session: https://claude.ai/code/session_01EFw2WdKr1oxvaKCJC2ua2R --- .claude/board/AGENT_LOG.md | 6 + .claude/board/LATEST_STATE.md | 6 + .claude/board/PR_ARC_INVENTORY.md | 42 +++++++ .../2026-10-16-ledger-revision-after-1576.md | 22 ++++ .claude/board/entries/README.md | 3 +- .../LEDGER-REVISION-2026-10-16.md | 105 ++++++++++++++++++ 6 files changed, 183 insertions(+), 1 deletion(-) create mode 100644 .claude/board/entries/2026-10-16-ledger-revision-after-1576.md create mode 100644 .claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16.md diff --git a/.claude/board/AGENT_LOG.md b/.claude/board/AGENT_LOG.md index 36e1f5780..b1644e71e 100644 --- a/.claude/board/AGENT_LOG.md +++ b/.claude/board/AGENT_LOG.md @@ -1,3 +1,9 @@ +## 2026-10-16 — ledger revision after #1576: three Haiku runs + +- `quote-harvester` (PR bodies): refused correctly, the card has no GitHub or Write tool. Re-run as general-purpose on Haiku: 16/16 fetched; 7 bodies read in full and used; 0 wrong. +- `harvest-scout` (band_reading callers, stale "two readings" prose): positive hits correct; 3 zero-hit claims false (`causal-plane-inventory.md:55`, `TYPE_DUPLICATION_MAP.md:15`, `causaledge64-mailbox-rename-soa-v1.md:57`). Second miss on zero-hit Markdown searches: that kind goes back to Sonnet. +- Cost: about 318k, 427k and 331k subagent tokens. No conclusion came from a card. + ## 2026-10-16 — /5plus3 council: metta-datalog-assimilation W1–W3 spec (D-ASM-1..4) - Phase 0: spec v1 by the orchestrator; premise-auditor (Opus) → PREMISE-SPLIT (stamp sub-premise PREMISE-WRONG); six of its code claims re-read and confirmed; spec re-asked as v1.1 before any savant was cast. diff --git a/.claude/board/LATEST_STATE.md b/.claude/board/LATEST_STATE.md index 192eafbb2..2daca5d84 100644 --- a/.claude/board/LATEST_STATE.md +++ b/.claude/board/LATEST_STATE.md @@ -1,3 +1,9 @@ +## 2026-10-16 — main after #1576 (merged `9a657bd3`); catch-up for #1560, #1570, #1571, #1574, #1575, #1576 + +- **On main, not recorded here before:** #1560 (O22 `CounterfactualVerdict` producer, R1 test, ledger correction), #1570 (Haspelmath harness, frozen operand dictionary), #1571 (D-QPROF-0 `qualia_profile`, merged; the "in PR" section below is now on main), #1574 (59..60 one field, two vocabularies; mapping inventory; D-ASM-0), #1575 (D-HOLO H14d measured), #1576 (D-ASM-1..4 spec ratified, nothing authorized). +- **Contract inventory delta:** `lance_graph_planner::pearl::{Reaction::verdict, Measured::counterfactual_verdict}` (#1560); `lance_graph_contract::qualia_profile` / `qualia_profile_table` (#1571, listed below). +- **Ledger:** `.claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16.md` maps #1560–#1576 onto the open items. + ## 2026-10-16 — main after #1573 (merged `461bbd1a`); D-HOLO H14c on main - **On `main` (#1573):** D-HOLO H14c probes and results (thought-first reads, Lance 13 masked I/O). Probe code only, no library change. diff --git a/.claude/board/PR_ARC_INVENTORY.md b/.claude/board/PR_ARC_INVENTORY.md index 7c21162c9..d4fb40598 100644 --- a/.claude/board/PR_ARC_INVENTORY.md +++ b/.claude/board/PR_ARC_INVENTORY.md @@ -1,3 +1,17 @@ +## 2026-10-16 — lance-graph PR #1576 (merged `9a657bd3`, branch `ccr-c09987ed-cuk6r5`, 8 commits) — council: metta-datalog-assimilation W1–W3 spec ratified (D-ASM-1..4) + +- **Added:** `.claude/plans/metta-datalog-assimilation-v1-council-v3.md` (executable spec, gates G1–G9); ⊘ notes on the plan's decisions A–E, G and §6; `metta-polyfill-v1.md` §5 amended; STATUS D-ASM-1..5, D-MTP-2, D-RA-3; 3 TECH_DEBT rows; entry `2026-10-16-metta-datalog-assimilation-council-v3.md`. +- **Locked:** premise gate returned PREMISE-SPLIT (a conjunctive body's existence half is quack `Filter::{Semijoin, EqU32Via}`; its answer half needs tuples with multiplicity) and the stamp sub-premise PREMISE-WRONG: a stamp is an evidence lane beside the atom, never inside it, with reserved value `3` = `NO_PROVENANCE`. Bag = keyed count `Terminal::GroupReduce{Count}`; new u16 `LabelCodebook`; route at `add`, refuse at `query`. +- **Deferred / open:** nothing authorized to build; D-ASM-1..4 Queued. Cross-table projections have no tuple carrier (W4+); `hyperon-probe/golden` is absent, so the 26-script baseline is measured before `LanceSpace`. +- **Confidence:** n/a (docs); council 49 findings, 0 BLOCK, 2 FIX-P1, 16 FIX-P2. + +## 2026-10-16 — lance-graph PR #1575 (merged `7b62af48`, branch `ccr-9f3b8075-74u47l-holo-h14d`, 8 commits) — D-HOLO H14d: logic → address mask → masked read → terminal (C only) + +- **Added:** `crates/holo-h14d-probe` (excluded from the workspace), plan `d-holo-h14d-logic-mask-read-v1.md` (pre-registered, amendments A1/A2), entry `2026-10-16-d-holo-h14d-logic-mask-read.md` with the raw CSV, STATUS row, `ISS-RELALG-ABSORPTION-AFTER-DE-MORGAN`, arc/LATEST entries for #1573. +- **Locked (v4):** relalg decides 675 of 2,000 programs symbolically (396 only after the rewrite), 1,065 vs 5,898 passes, 1.9 µs per program; the address mask costs 2–10 ns per 64-address word; Lance rows scanned equal |R| in every cell, but bytes are page-bound (uniform 1 % reads 1.00 of the bytes, clustered 1 % reads 0.19); about 6 ns per lane load; point query 8–45 ns hot. +- **Deferred / open:** page-local clustering; the payload stage scans every word of R (fixed cost, prediction 4 falsified); 2 of 2,000 programs widened by `normalize`. +- **Confidence:** medium; checks 1–8 pass, five disable runs red; one VM. No layout decision depends on these numbers. + ## 2026-10-16 — lance-graph PR #1573 (merged `461bbd1a`, branch `ccr-9f3b8075-74u47l-holo-h14c`, head `e4681f10`, 11 commits) — D-HOLO H14c: thought-first reads, Hamming presence, Lance 13 masked I/O - **Added:** `holo_thought_first_probe` (point projection, mask-risc populations, terminal rows; 10 oracle tests) and the `lane_footprint` instrument in `lance-graph-mask-risc/examples`; `crates/holo-lance-masked-io-probe` (excluded from the workspace: Lance 13 masked reads, stale-map hazard across rewrites); plan `d-holo-h14c-thought-first-v1`, entry `2026-10-15-d-holo-h14c-thought-first.md`, STATUS row D-HOLO-H14c. @@ -5,6 +19,27 @@ - **Deferred / open:** the C vs B comparison lines in that PR rest on the withdrawn row-first framing; C's own four-stage operation is D-HOLO-H14d (#1575). Positional tables are page-dense at uniform 1 % (≈ 93× payload resident). - **Confidence:** medium; oracle tests and disable runs red; about 40 % run-to-run noise on the VM. +## 2026-10-16 — lance-graph PR #1574 (merged `4f9d5f26`, branch `ccr-c09987ed-cuk6r5`, 6 commits) — CE64 vocabulary correction, NARS/MeTTa mapping inventory, metta-datalog-assimilation plan + +- **Added:** research `2026-10-15-nars-metta-ce64-mapping-inventory.md` (read from hyperon-experimental `3f76dc4`, PeTTa `1d19a1d`, ONA `dc4efd0`, opennars `4d1bdb0`, Narsese.rs `16d4d18`); plan `metta-datalog-assimilation-v1.md` (D-ASM-0..7); three entries; doc corrections in `causal-edge/src/{layout,lib}.rs`, `contract/src/{band_reading,epistemic_state5}.rs`. +- **Locked:** bits 59..60 are one field under two vocabularies (`TrustTexture` Crystalline…Murky ≡ `CausalTopology` Direct…Unknown, identity mapping); contradiction is an epistemic pothole the MUL reacts to, not a cell; only `ReasoningBand` on 61..63 wrote another meaning. No foreign system carries a 59..63 or a 40..42; lance-graph's abduction/induction argument order matches OpenNARS Java, ONA and PeTTa swap it; evidential-base width differs per engine (`Stamp(u64)`, `&[u64]`, ONA 10, OpenNARS 20,000). +- **Deferred / open:** relalg join + `μ` is the most unblocking gap; the per-row truth law is the one candidate byte; stale prose listed in the PR body (`causal-plane-inventory.md:55`, `TYPE_DUPLICATION_MAP.md:15`, two historical plans); `project_truth` still refuses the other vocabulary. +- **Confidence:** high for the doc correction (executable contract unchanged, tests green); the inventory is read-from-source, not executed. + +## 2026-10-16 — lance-graph PR #1571 (merged `4d57fe20`, branch `qualia-profile-icc`, 3 commits) — contract: `qualia_profile`, an ICC-style named-qualia table (D-QPROF-0) + +- **Added:** `data/qualia/calibrated_219.{json,bin1024}`, `scripts/gen_qualia_profile.py`, `lance_graph_contract::{qualia_profile, qualia_profile_table}` (219 named patches, one-byte palette `0 | 1..=219 | 220..=255 reserved`, `nearest_named`, `Gamut`, `Mix`, `device_to_pcs_shared`); `ISS-QUALIA-A2B-CLUT-UNMEASURED`. +- **Locked:** three spaces kept apart (device, PCS, observer); only the five same-named axes map device → PCS, returned with a `measured` mask; the table is pinned by `PROFILE_ID` and `TABLE_FNV`. +- **Deferred / open:** the CLUT for the other twelve axes (a chart probe); how 219 named patches share the Σ-codebook byte with the 64 `sigma_rosetta` glyphs. +- **Confidence:** high for the table and its 11 tests (FNV mutation red); the same-name hypothesis is stated as a hypothesis. + +## 2026-10-16 — lance-graph PR #1570 (merged `aaab596c`, branch `research/haspelmath-deepnsm-ce64-hydrate-harness`, 10 commits) — research(deepnsm-v2): Haspelmath grammar harness on a frozen 8:8:8 operand ABI + +- **Added:** `deepnsm-v2/src/haspelmath.rs` (Fig. 8.1 adjacency, connectedness falsifier, token binder, a frozen epoch-wide `u16 WordId ↔ u8` operand dictionary with version check, six tests), example `haspelmath_wiring`, `.arxiv/2026-10_haspelmath-deepnsm-ce64-harness.md`. +- **Locked:** the 8:8:8 lanes are one register contract; the per-verse palette of an earlier draft was removed. CE64's upper 40 bits stay zero: grammar mints no confidence, Pearl operation, maturity or source root. +- **Deferred / open:** the dictionary is a fixture until a canonical cross-system registry exists; TEKAMOLO × case × anaphora, a source-backed writer, Bible-scale and held-out UD accuracy are held. +- **Confidence:** low until run: the body states the tests were committed but not run locally. + ## 2026-10-16 — lance-graph PR #1569 (merged `09f31561`, branch `ccr-9f3b8075-74u47l-holo-h14b`, head `2f07941c`) — D-HOLO H14b: lane mask M measured, then withdrawn - **Added:** arm M (C's per-lane tables plus one `u32` per row naming the changed lanes) in `holo_lane_storage_probe` with 4 tests; plan section, entry `2026-10-15-d-holo-h14b-lane-mask.md`, STATUS row D-HOLO-H14b (WITHDRAWN). @@ -47,6 +82,13 @@ - **Deferred / open:** the 60-minute soak; seeding the version clock from storage on open; compaction/deletes break the clock's prefix property (fragment-id remap needed); clock not wired into a writer or reader. - **Confidence:** high for V1/V2 and the cache cap (test-first, disable runs red); medium for D-WAL-2/3 timings (one VM, soak truncated at 27.5 min). +## 2026-10-15 — lance-graph PR #1560 (merged `4f47f4bc`, branch `ccr-b2e415d9-4jfvyk`) — O22 counterfactual verdict, R1 pin, ledger correction (CE64 as one wire contract) + +- **Added:** `pearl::Reaction::verdict(factual_consistent)` and `Measured::counterfactual_verdict`; `kanban` R1 test; `.claude/research/2026-10-recovery/LEDGER-CORRECTION-2026-10-15.md`; entry `2026-10-15-ledger-correction-wire-contract.md`; 11 arc entries for #1494–#1504; STATUS rows D-RESIDUE-FP-0, D-MATURATION3-0/-1. +- **Locked:** the first end-to-end `RevisionVerdict::is_acceptable()` over a real replay; only a load-bearing cut from a consistent chain is `Necessary` (a cut that repairs an inconsistent chain is `Dispensable`, Codex P1). Rulings R-a..R-f: CE64 is one global wire; 40..42 is Pearl's ladder as 50..52 is the pyramid; O25 is a defect with a known fix. +- **Deferred / open:** O23 (TruthOnly reading), ablation (defined later through the N15 families), the 46..49 meaning, the derived-packet rule, the migration gate. +- **Confidence:** high; 6 disable runs red. + ## 2026-10-14 — lance-graph PR #1555 (merged `5366a5ab`, branch `ccr-9f3b8075-74u47l-metta`, head `457a8c41`, 14 commits) — D-MTP: MeTTa over the masking-algebra polyfill; two-hop join and the hop as a range - **Added:** plan `metta-polyfill-v1.md` (Hyperon seams, PeTTa vs Hyperon measurements, MeTTa-as-IR verdict, datafrog fork as the exact Rust oracle). `crates/lance-graph-metta` (own workspace, excluded; datafrog from `AdaWorldAPI/datafrog` as a dev-dependency only) with `join_probe` (D-MTP-1, two-hop join through Quack `Semijoin`) and `join_range_probe` (D-MTP-1b, the hop as a range under a monotone `dst`). Entries `2026-10-14-metta-polyfill-map.md`, `-d-mtp-1-join.md`, `-d-mtp-1b-range-hop.md`; STATUS D-MTP rows. diff --git a/.claude/board/entries/2026-10-16-ledger-revision-after-1576.md b/.claude/board/entries/2026-10-16-ledger-revision-after-1576.md new file mode 100644 index 000000000..77241e64a --- /dev/null +++ b/.claude/board/entries/2026-10-16-ledger-revision-after-1576.md @@ -0,0 +1,22 @@ +# 2026-10-16 — Recovery ledger revised against #1560–#1576 + +**Status:** WORKING-MODEL (research document; board catch-up; no code changed). + +`.claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16.md` maps 15 +merged PRs and two closed drafts onto the 2026-10-15 ledger correction. + +- **Closed:** C22 O22 verdict producer (#1560); C23 R1 over the full grid + (#1565); C24 bits 59..60 one field, two vocabularies (#1574); C25 one frozen + operand dictionary per epoch (#1570); C26 N15 mapping inventory from source + (#1574); C27 stamp = evidence lane beside the atom (#1576); C28 version clock + (#1562, #1567); C29 no per-predicate `hydrate` guard (#1572 rejected); C30 + arc entries for #1560, #1570, #1571, #1574, #1575, #1576. +- **Merged:** O25 + K2 into one evidence-root item; N1 rest + derived-packet + decision into D-HL-5 H4 (pre-registered in #1564); ablation + N15 + D-ASM + into one stream; the README tension + #1574's stale prose into DOC-WIRE. +- **New:** O29 (accepted verdict mid-action), O30 (evidential-base width), + BAND-RETIRE, N16 operand registry, N17 relalg absorption, N18 clock reader, + N19 layout costs, N20 qualia CLUT, N21 run the Haspelmath tests. + +OPEN: O30, O29, DOC-WIRE (1), BAND-RETIRE, D-ASM authorization (decisions); +O25 fix and D-HL-5 H4 (high fruits, ready once O30 is decided). diff --git a/.claude/board/entries/README.md b/.claude/board/entries/README.md index f4e1b09ed..4afcdf9fe 100644 --- a/.claude/board/entries/README.md +++ b/.claude/board/entries/README.md @@ -25,11 +25,12 @@ 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. -350 entries, 2026-08-06 .. 2026-10-16. +351 entries, 2026-08-06 .. 2026-10-16. | date | entry id | finding | file | |---|---|---|---| | 2026-10-16 | `D-ASM-1` | | [2026-10-16-metta-datalog-assimilation-council-v3.md](2026-10-16-metta-datalog-assimilation-council-v3.md) | +| 2026-10-16 | `ledger-revision-after-1576` | | [2026-10-16-ledger-revision-after-1576.md](2026-10-16-ledger-revision-after-1576.md) | | 2026-10-16 | `D-HOLO-H14d` | | [2026-10-16-d-holo-h14d-logic-mask-read.md](2026-10-16-d-holo-h14d-logic-mask-read.md) | | 2026-10-15 | `nars-metta-ce64-mapping-inventory` | | [2026-10-15-nars-metta-ce64-mapping-inventory.md](2026-10-15-nars-metta-ce64-mapping-inventory.md) | | 2026-10-15 | `D-ASM-0` | | [2026-10-15-metta-datalog-assimilation-plan.md](2026-10-15-metta-datalog-assimilation-plan.md) | diff --git a/.claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16.md b/.claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16.md new file mode 100644 index 000000000..cfba0fafb --- /dev/null +++ b/.claude/research/2026-10-recovery/LEDGER-REVISION-2026-10-16.md @@ -0,0 +1,105 @@ +# Ledger revision, 2026-10-16 + +Maps the PRs since the last mapping (#1560–#1576) onto +`LEDGER-CORRECTION-2026-10-15.md` (merged in #1560). That file and the +2026-10-14 revision are not edited; this file supersedes the rows it names. +Board entry: `.claude/board/entries/2026-10-16-ledger-revision-after-1576.md`. + +**Scope.** 15 merged PRs (#1560, #1562–#1571, #1573–#1576) and two closed +unmerged drafts (#1561 withdrawn, #1572 rejected). Read from the merged PR +bodies (fetched verbatim), the code diffs of #1562, #1565 and #1574, the board +entries each PR added, and the arc entries. + +## 1. New closed ends + +| id | what closed | evidence | +|---|---|---| +| C22 | O22, the first `CounterfactualVerdict` producer, including the flip direction (a cut that repairs an inconsistent chain is `Dispensable`) | #1560; `pearl.rs` tests `o22_*`; 6 disable runs red | +| C23 | R1 covers the whole verdict grid: all nine (effect × counterfactual) verdicts from `CognitiveWork`; none reopens the goal; the one accepted verdict moves on to `Evaluation`, which is not a goal revision | #1565 (strengthens the #1560 test) | +| C24 | Bits 59..60 are one field under two vocabularies (`TrustTexture` ≡ `CausalTopology`, identity mapping); contradiction is a pothole the MUL reacts to, not a cell | #1574 code-doc correction in `layout.rs`, `band_reading.rs`, `epistemic_state5.rs` | +| C25 | The global-wire ruling (R-b) holds on 0..23 too: the Haspelmath harness replaced a per-verse operand palette with one frozen, version-identified dictionary per epoch | #1570 | +| C26 | N15 groundwork read from source: no foreign system carries a 59..63 or a 40..42 (both are ours to write); abduction/induction argument order matches OpenNARS Java, ONA and PeTTa swap it; evidential-base width differs per engine | #1574 mapping inventory (pinned SHAs) | +| C27 | The stamp's place: an evidence lane beside the atom, never inside it, reserved `3` = `NO_PROVENANCE` | #1576 council (premise-auditor PREMISE-WRONG on "stamp as atom") | +| C28 | The coordinate machinery the O25 fix and K2 need exists: `VersionClock` maps a Lance version to the row positions it holds; the writer keeps it current | #1562 (contract), #1567 (writer) | +| C29 | A per-predicate guard on `hydrate` is not needed: a chain handed to `hydrate` is one relation's chain by construction, and `ChainStep.0` is not the predicate | #1572 rejected and reverted; closeout in its body | +| C30 | Board catch-up: arc entries for #1560, #1570, #1571, #1574, #1575, #1576; LATEST_STATE section | this PR | + +## 2. Open items, revised (merged, deduplicated) + +Fruit: **L** low (an afternoon, no decision needed), **H** high (a day or a +measured probe), **Heavy** (multi-PR or blocked on a decision). Tier is who +should do the grindwork. + +| id | item | status after #1576 | fruit | tier | +|---|---|---|---|---| +| **O25 + K2** (merged) | Replay revises at every hop. Fix: compose along one trajectory; revise only across distinct evidence roots. The root is the `temporal.rs` coordinate, resolvable through `VersionClock` (C28) and carried as an evidence lane (C27). | OPEN, now fully specified. Two pinned tests in `chain_confidence.rs` flip with a recorded reason. D-HL-5 H5 (NAL overlap vs Gadamer gate, pre-registered in #1564) is the paired check. | H | O designs the root type, S implements, guarded-executor gate | +| **O30** (new, from C26) | Evidential-base width: `Stamp(u64)` bitset of ≤ 64 sources (planner), `&[u64]` roots (nal), ONA 10, OpenNARS 20,000. Pin the policy before revision runs over large populations. | OPEN (decision) | — | O | +| **K6 = N1 rest + decision 2** (merged) | What a derived conclusion writes into 53..63. #1564 pre-registered it as D-HL-5 H4 with candidate laws (a) 0, (b) `IndirectKnown × Open`, (c) weakest premise topology made indirect × `Open`, (d) `Unknown × Open` then hydrated, and a decision rule. | OPEN, experiment written, not run | H | S runs the lane, O reads the table, operator picks | +| **O29** (new, from #1565) | Should an accepted verdict move `CognitiveWork` on mid-action, or wait for the action to finish? | OPEN (decision) | — | O | +| **D-46** | One global meaning for 46..49: opcode (code) or activation (design). | unchanged | Heavy | O | +| **D-MATURATION3-1** | The pyramid as the plasticity ladder in 50..52: entry code (7 vs "new"), `learn`'s rule, migration gate for codes 1..6 (`learn`, `MooreNars16`, `CausalEdgeV3` byte 3 in lockstep). | unchanged | Heavy | O, then S behind a version gate | +| **ABL** (item 6) | Ablation through the N15 families. Its machinery now exists: D-ASM §7 "validate against the original as the oracle first" and the pinned upstream SHAs (C26). | OPEN; a `/coresearch` run is the next step | H | coresearch council | +| **N15 → D-ASM** (merged) | Pearl × NAL × others as views on one wire. Executable path: D-ASM-1..4 (binder, join as preimage, bag terminal, `LanceSpace` + three-way harness), ratified in #1576, nothing authorized. First step: measure the 26-script Hyperon baseline (`hyperon-probe/golden` is absent). | Queued (authorization) | Heavy | O authorizes; S builds W1 | +| **O23** | TruthOnly: a fourth verdict or a remap. Caveat unchanged: at the verdict level TruthOnly and "no cut" both read `NotRun`. | unchanged | — | O | +| **R2–R4** | Rubikon tests: agreement as `HorizonFusion` (needs "independent root for a person" decided), goal never revised by counting, folding only under the gate. | unchanged | L for R3/R4 once worded; R2 blocked | S | +| **O26** | `causal_bridge` Intentional → Causes, dead code behind a non-compiling `wip` feature. | unchanged (downgraded) | — | — | +| **O7, O13, O15** | model-identifier rule, phase/activation rule, Experiment A. | unchanged | — | O | +| **DOC-WIRE** (merged: README tension + #1574 leftovers) | (1) `docs/architecture/ce64-semantic-upper-half.md` lets each tenant declare its own reading of 43..52; R-b says one global meaning. (2) Stale prose still saying "two readings" on 59..60: `.claude/knowledge/causal-plane-inventory.md:55`, `docs/TYPE_DUPLICATION_MAP.md:15` ("Pearl-3 epistemic lens"), `.claude/plans/causaledge64-mailbox-rename-soa-v1.md:57` (`Murky/Dissonant … Contradiction`). | OPEN; (1) awaits a yes; (2) is mechanical | L | spec-editor (Haiku) per file, O reads each diff | +| **BAND-RETIRE** (new) | `band_reading` has no production caller (only the `ClassView::band_reading` trait method at `class_view.rs:1338` and two examples); `project_truth` still refuses the other vocabulary with `LensMismatch`, a gate on a distinction C24 removed. Either retire the 59..60 half of the declaration surface or keep it for 61..63 `ReasoningBand` only. | OPEN (small decision) | L/H | S after the decision | +| **N16** (new, from #1570) | A canonical, cross-system operand registry: `u16 WordId → u8` operand per sealed epoch. Until it exists every 8:8:8 operand dictionary is a fixture, and C25 is a rule without a registry. | OPEN | Heavy | O designs | +| **N17** (new, from #1575) | `ISS-RELALG-ABSORPTION-AFTER-DE-MORGAN`: `normalize` widens 2 of 2,000 programs (absorption missed after De Morgan). Answers stay correct; the cost is width and a refusal. | OPEN, reproducer in the H14d CSV | L/H | S with the two programs as fixtures | +| **N18** (new, from #1562/#1567) | The version clock is not wired into a reader; compaction and deletes break its prefix property (fragment-id remap); the clock is not seeded from storage on open; the 60-minute soak is outstanding. | OPEN | H | S | +| **N19** (new, from #1564/#1566–#1575) | Layout stream: no A/B/C layout verdict yet. Open costs: the merged point read (~21 overlay entries per row, 2.3 µs), page-bound masked reads (bytes saved only when survivors are page-local), the payload stage's fixed word scan. | OPEN, measured, no decision depends on it yet | Heavy | O | +| **N20** (new, from #1571) | `ISS-QUALIA-A2B-CLUT-UNMEASURED`: the device → PCS mapping for 12 of 17 axes; the Σ-codebook slot sharing between 219 named patches and 64 glyphs. | OPEN | H | S runs the chart probe | +| **N21** (new, from #1570) | The Haspelmath harness's tests were committed without a local run. | OPEN, verify | L | guarded-executor (Haiku) runs the five commands in the PR body | + +## 3. Connecting the dots + +- **K2' — one evidence root.** C27 (stamp as an evidence lane), C28 (version → + row positions) and the O25 fix are one object: an evidence root is a + (lane, row position) pair, resolvable as of any version, carried beside the + atom. O30 (width) is the only decision left before it can be built. Once it + exists, `nars::belief::Stamp`, `nal::Evidence.stamp`, contract + `independent_roots` and the probes' roots key on it, replay stops pooling + self-evidence, and D-HL-5 H5 can be read. +- **K6 — the derived packet is an experiment, not a debate.** #1564 already + wrote the four candidate laws and the rule that picks between them. Running + H4 answers decision 2 and the rest of N1 with numbers. +- **K7 — ablation, N15 and D-ASM are one stream.** The families the operator + named for ablation are exactly D-ASM's oracles (Hyperon, PeTTa, ONA, + OpenNARS), with the SHAs already pinned. A `/coresearch` run on "what does + removing one piece of evidence mean in each family" uses that inventory + rather than re-reading upstream. +- **K8 — the wire is now global in practice, not just in prose.** C24 (59..60), + C25 (0..23) and R-b agree. What still says otherwise is documentation + (DOC-WIRE) and one dormant gate (BAND-RETIRE). + +## 4. Fruits + +- **Low:** DOC-WIRE (2) — three stale lines; N21 — run five commands; N17 — + two reproducers become tests, then the absorption rule; R3/R4 wording. +- **High:** O25 + K2' once O30 is decided; D-HL-5 H4 (K6); the ablation + `/coresearch`; N18 reader wiring; N20 chart probe. +- **Heavy:** D-ASM W1 (authorization); D-MATURATION3-1 migration; D-46; N16 + registry; N19 layout verdict. + +## 5. Decisions for the operator + +| # | question | recommendation | +|---|---|---| +| O30 | evidential-base width | a bounded root list keyed on (lane, row position), width pinned per class, overflow refused rather than truncated | +| O29 | accepted verdict mid-action | wait for the action to finish; record the verdict, move at the boundary | +| DOC-WIRE (1) | amend the diagram doc to one global reading of 43..52 | yes, with the version-gate sentence kept | +| BAND-RETIRE | retire the 59..60 lens gate | keep the declaration for 61..63 only; drop `TruthLens` refusal | +| D-ASM | authorize W1 | yes for the Hyperon baseline measurement only | + +## 6. Agent ledger (Haiku harvest tier) + +| card | runs | items checked | wrong | missed | +|---|---|---|---|---| +| quote-harvester (PR bodies) | 1 | — | — | refused correctly: the card has no GitHub or Write tool | +| general-purpose on Haiku (PR bodies #1561–#1576) | 1 | 7 bodies read in full and used; titles/merged state cross-checked against `git log` | 0 | 0 | +| harvest-scout (band_reading callers, drift lines) | 1 | positives `class_view.rs:1338`, `band_reading.rs` sites; 3 zero-hit claims re-checked | 0 | **3** (patterns 5, 6, 8 present at the cited lines) | + +Two misses on zero-hit searches over Markdown (one 2026-10-14, three here): +that kind of task goes back to Sonnet. Positive-hit harvesting on Haiku stays.