diff --git a/.claude/board/AGENT_LOG.md b/.claude/board/AGENT_LOG.md index 8cd9e124..2090b90f 100644 --- a/.claude/board/AGENT_LOG.md +++ b/.claude/board/AGENT_LOG.md @@ -1,3 +1,10 @@ +## 2026-07-15 — cross-repo forensic audit (q2 session) — rung-2 two-144s split + third 0–9 ladder filed as deltas onto E-RUNG-CONTENT-LADDER-1 + +- **Fleet:** 5-agent read-only evidence sweep (before/after `1a11038` API shape, contract StyleFamily design, D-TSC-1 spec/governance coverage, five-tables divergence reconstruction, cross-repo consumer+CI survey) + main-thread verification greps. A second 4-agent sweep (per-hunk claim audit) was lost to a token wall — its two open checks (FieldModulation→plan-shape functional trace; full 38-file claim audit) remain UNRUN, not asserted. +- **Dedup discipline:** audit initially ran against pre-ruling `757c3e8`; after fetching main, findings were deduped against `E-RUNG-CONTENT-LADDER-1`, `E-COMPAT-ALIAS-MUST-BE-LIFTED-1`, TD-STYLE-TABLE-RESIDUE item-3 probe, and the existing RungLevel byte-dup TD paragraph — only two genuinely unfiled findings shipped. +- **Deliverables (doc-only, this commit):** `EPIPHANIES.md` `E-RUNG2-TWO-144S-1` prepended; `TECH_DEBT.md` `TD-RUNG2-144-VOCAB-SPLIT` + `TD-THIRD-RUNG-LADDER-LEARNING` prepended; `persona-vs-rung-ladder.md` O7/O8 appended. +- **Outcome:** rung-2 wiring (O1) now has its gating adjudication named; no ruling changed, no code changed. + ## 2026-07-14 — 5+3 council: OGAR-vs-lance-graph layer-confusion audit + contract doc-label cleanup - **Deliverable:** doc-comment label-bleed removed from `contract::class_view::{execute_compute_dag, execute_defaults}` (were branded "the ActionDef value executor" + carried MedCare/sono provenance in the zero-dep contract). E-LAYER-CONFUSION-OGAR-VS-SPINE-1 prepended. New issue #692 (contract::action shape-parity fuse gap). - **Council:** 5 falsification savants (Sonnet ×4 / Opus ×1) + 3 brutal reviewers (Sonnet). Grades: 3 WRONG, 1 UNSOUND, 1 WRONG-not-catastrophic. Reviewers raised 2 BLOCK + 6 FIX against the orchestrator's own remediation (all accepted). #690 placement verified CORRECT (not reverted). diff --git a/.claude/board/EPIPHANIES.md b/.claude/board/EPIPHANIES.md index b15090cc..ce822dc4 100644 --- a/.claude/board/EPIPHANIES.md +++ b/.claude/board/EPIPHANIES.md @@ -21,6 +21,20 @@ The D-QUANTGATE rationale restatement: the integer walk is canon for the quantiz **Load-bearing consequences for THIS repo:** (1) blasgraph is re-affirmed as the bit-exact canonical kernel home — the f32 tropical-GEMM kernel is UNWRITTEN; `bgz17::ScalarCsr::spmv_min_plus` is the only shipped min-plus and is a lossy sibling, never the canon (audit ground-truth #3/#5). (2) NNVC-in-the-decode-loop is the public antithesis of the workspace anti-neural rule ("NNs train tables, never run in the hot loop") — determinism is the moat for evidentiary/medical/scientific video, and mobile-vendor decoder-complexity pushback is evidence for our side. (3) ECM's tool-accretion complexity spiral is the external cautionary anchor for the R-3 LoC envelope. Watch dates: Nov 2026, Jan 2027. **Cross-ref:** ndarray `pr-x12-h266-h267-standards-landscape.md` (sources), ndarray `PR-X12-docs-audit.md` §8 addendum (what was applied), `3DGS-PRX12-crosspollination-capstone.md` standards-watch section (this commit). +## 2026-07-15 — E-RUNG2-TWO-144S-1 — rung 2's "144 verbs" are TWO divergent vocabularies (sigma_rosetta vs verb_table); and a third 0–9 ladder (`learning::Rung`) sits off-canon + +**Status:** FINDING (file:line receipts, verified on main; cross-repo audit session, initially run against pre-ruling `757c3e8`, deltas re-verified after `E-RUNG-CONTENT-LADDER-1` landed). This entry files ONLY what the 2026-07-14 ruling + TD ledger do NOT already carry — it does not restate the ladder, the persona demarcation, or the contract↔engine `RungLevel` byte-dup (already TD-ledgered). + +**Finding 1 — the two 144s of rung 2 disagree.** `E-RUNG-CONTENT-LADDER-1`'s rung-2 row cites `sigma_rosetta.rs` + `grammar/verb_table.rs` jointly as "the 144 thinking atoms," but the two tables are divergent hand-rolled vocabularies: +- **Root/family axis ~2/12 overlap:** sigma `VERB_ROOTS` = sense, feel, think, act, become, hold, release, create, transform, connect, separate, integrate (`sigma_rosetta.rs:40-53`) vs contract `VerbFamily` = Becomes, Causes, Supports, Contradicts, Refines, Grounds, Abstracts, Enables, Prevents, Transforms, Mirrors, Dissolves (`verb_table.rs:48-61`). Only become/Becomes + transform/Transforms align. +- **Tense axis differs in membership AND ordinals:** sigma has one "continuous" + explicit "subjunctive" + "gerund" (`sigma_rosetta.rs:55-68`); contract `Tense` has THREE continuous variants, no subjunctive/gerund, `Potential` documented as the Subjunctive stand-in (`role_keys.rs:244-257`). Ordinal skew on the shared names: habitual = 7 (sigma) vs 9 (contract); potential = 8 vs 10; imperative = 9 vs 11. +- **Both are ordinal-indexed** (`root % 12` / `tense % 12` at `sigma_rosetta.rs:78-83`; `family as usize`/`tense as usize` at `verb_table.rs:198-200`, with verb_table's own doc warning that reordering is load-bearing). A naive rung-2 bridge between them would silently mis-map — the pre-M9 five-tables signature, one layer down. +- **Consequence:** O1 (rung↔content wiring) is gated on an adjudication this finding surfaces: are the two 144s one atom vocabulary + its parse-side projection (then an explicit, parity-tested mapping is required), or two KINDS (then the demarcation doc needs the split named, as it did for persona-36)? Filed as `TD-RUNG2-144-VOCAB-SPLIT`. + +**Finding 2 — a THIRD 0–9 ladder exists off-canon.** `learning::cognitive_frameworks::Rung` (`cognitive_frameworks.rs:699-713`): Noise=0, Token, Phrase, Proposition, Argument, Narrative, Worldview, Paradigm, Episteme, Transcendent=9 — a meaning-granularity ladder sharing the 0–9 shape and the `Transcendent=9` endpoint with `RungLevel`, but with different interior semantics (ordinal 4 = "reasoning chain" there vs "generalized patterns" in RungLevel). Endpoints deceptively agree; every interior rung diverges. The TD ledger covers only the contract↔engine byte-dup pair; this third ladder has no row and no demarcation. Adjudication needed: separate storyline (à la persona-36) or rename so "rung" stops meaning two ladders. Filed as `TD-THIRD-RUNG-LADDER-LEARNING`. + +**Cross-refs:** `E-RUNG-CONTENT-LADDER-1` (the ruling this audits under), `.claude/v3/knowledge/persona-vs-rung-ladder.md` (O7/O8 added), `TD-RUNG2-144-VOCAB-SPLIT`, `TD-THIRD-RUNG-LADDER-LEARNING`, the existing RungLevel byte-dup TD paragraph (2026-07-x, "thinking-engine does not dep lance-graph-contract"), MODULE-TABLE rows for `verb_table.rs` (:184) and `sigma_rosetta.rs` (:467) — censused as unrelated rows, which is how the split stayed invisible. + ## 2026-07-14 — E-LAYER-CONFUSION-OGAR-VS-SPINE-1 — a session-wide OGAR-vs-lance-graph layer conflation, caught by a 5+3 council; contract doc-comment label-bleed cleaned **Status:** FINDING (5+3 council, source-receipted; MedCare-rs E-MEDCARE-30 carries the full grades). diff --git a/.claude/board/TECH_DEBT.md b/.claude/board/TECH_DEBT.md index ed90d6a5..f934eca5 100644 --- a/.claude/board/TECH_DEBT.md +++ b/.claude/board/TECH_DEBT.md @@ -1,5 +1,44 @@ # Technical Debt Log — Open + Paid (double-entry, append-only) +## TD-RUNG2-144-VOCAB-SPLIT — the two 144s of rung 2 are divergent vocabularies (2026-07-15) + +**Open.** `E-RUNG-CONTENT-LADDER-1`'s rung-2 row cites `sigma_rosetta.rs` +(`VERB_ROOTS`/`VERB_TENSES`, the wired-hot-path Membrane atoms) and contract +`grammar/verb_table.rs` (`VerbFamily`×`Tense` → TEKAMOLO `SlotPrior`, the +parse-side priors) jointly as "the 144 thinking atoms" — but the two tables +share only ~2/12 roots (become, transform), differ in tense MEMBERSHIP +(subjunctive+gerund+1×continuous vs 3×continuous, no gerund, Potential as +Subjunctive stand-in) and in tense ORDINALS on the shared names (habitual +7 vs 9, potential 8 vs 10, imperative 9 vs 11). Both are ordinal-indexed +(`root % 12` / `tense as usize`), so any naive rung-2 bridge silently +mis-maps — the pre-M9 five-tables signature one layer down. **This gates +O1** (rung↔content wiring) in `persona-vs-rung-ladder.md`. **Pay by** one +of: (a) unify on one vocabulary (heavy — sigma is a baked wire format); +(b) an explicit sigma↔verb_table mapping table in contract with a +literal-pin parity test on both axes (the D-TSC-1 treatment); (c) operator +demarcation "two KINDS" (atom vocabulary vs parse projection) recorded in +the demarcation doc + a doc-comment cross-warning at BOTH tables that the +other exists and ordinals do NOT transfer. Receipts + full diff: +`E-RUNG2-TWO-144S-1`. + +## TD-THIRD-RUNG-LADDER-LEARNING — `learning::Rung` is a third, semantically divergent 0–9 ladder (2026-07-15) + +**Open.** `learning/src/cognitive_frameworks.rs:699-713` defines +`pub enum Rung` — Noise=0, Token, Phrase, Proposition, Argument, Narrative, +Worldview, Paradigm, Episteme, Transcendent=9 — a meaning-granularity +ladder sharing the 0–9 shape and the `Transcendent=9` endpoint with +`contract::cognitive_shader::RungLevel`, while EVERY interior ordinal +means something different (4 = "reasoning chain" vs "generalized +patterns"). Endpoints deceptively agree; interiors diverge. The existing +RungLevel byte-dup entry (below) covers only the contract↔engine verbatim +pair — this third ladder has no row and no demarcation, and any code +exchanging rung ordinals with `learning` mis-means rungs 1–8 silently. +**Pay by** operator adjudication: (a) demarcate as a separate storyline +(à la persona-36 in `persona-vs-rung-ladder.md`) with a doc-comment +cross-warning, or (b) rename (`MeaningGrain`?) so "rung" stops meaning two +ladders. Cross-ref: `E-RUNG2-TWO-144S-1` Finding 2; +`E-RUNG-CONTENT-LADDER-1`. + ## TD-STYLE-TABLE-RESIDUE item 3 — PROBED, downgraded (2026-07-14) The "does any caller feed StyleFamily ordinals into p64-bridge diff --git a/.claude/v3/knowledge/persona-vs-rung-ladder.md b/.claude/v3/knowledge/persona-vs-rung-ladder.md index dc5024b2..8f88561f 100644 --- a/.claude/v3/knowledge/persona-vs-rung-ladder.md +++ b/.claude/v3/knowledge/persona-vs-rung-ladder.md @@ -103,6 +103,23 @@ its "runbook" naming and the board label cross storylines. macro object. `FUTURE-DESIGN.md` E-THINKING-STYLES-ARE-CLASSES-1 ("style = class: StepMask × WideFieldMask + rung set + KausalSpec") is the landing zone. +- **O7 — the two 144s of rung 2 are divergent (added 2026-07-15, + cross-repo audit; gates O1).** The rung-2 row above cites + `sigma_rosetta` + `verb_table` jointly, but their vocabularies + disagree: ~2/12 root/family overlap (become, transform), different + tense membership (subjunctive/gerund vs 3×continuous + + Potential-as-Subjunctive), and skewed ordinals on shared tense names + (habitual 7 vs 9, imperative 9 vs 11). Both ordinal-indexed — a naive + bridge silently mis-maps. Rung-2 wiring needs either a parity-tested + mapping or an explicit "atom vocabulary vs parse projection" + demarcation first. Receipts: `E-RUNG2-TWO-144S-1`; debt row: + `TD-RUNG2-144-VOCAB-SPLIT`. +- **O8 — a third 0–9 ladder, off-canon (added 2026-07-15).** + `learning::cognitive_frameworks::Rung` (Noise=0 … Transcendent=9) is a + meaning-granularity ladder whose interior ordinals all diverge from + `RungLevel` while the endpoints deceptively agree. Needs the persona-36 + treatment: demarcate as its own storyline or rename. Debt row: + `TD-THIRD-RUNG-LADDER-LEARNING`. ## Cross-refs