Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
35 commits
Select commit Hold shift + click to select a range
288b97f
deepnsm-v2: position decides noun/verb homographs, not the lemma tag …
claude Oct 4, 2026
d129744
deepnsm-v2: adverb tag, measured typology, and the position rules sco…
claude Oct 4, 2026
31c6736
deepnsm-v2: KJV silver labels from Luther 1545, clause rule as a swit…
claude Oct 4, 2026
ea7c077
deepnsm-v2: learned position table, German case kept, Animal Farm ref…
claude Oct 4, 2026
835c53e
deepnsm-v2: German TEKAMOLO cues by verb position; adverbial adjectiv…
claude Oct 4, 2026
9e8a552
deepnsm-v2: the question test as the Pearl 2³ mask; blast radius of d…
claude Oct 4, 2026
ec65f27
deepnsm-v2: literal rules, priorities and a quorum for word-class hom…
claude Oct 4, 2026
90c3750
deepnsm-v2: the Frageprobe through the Satzklammer; question mask is …
claude Oct 4, 2026
488243e
deepnsm-v2: a floating quantifier keeps the verb slot open (Bugbot, #…
claude Oct 4, 2026
e717e3d
deepnsm(-v2): restore the deleted grammar heuristics; WordNet identit…
claude Oct 4, 2026
558e802
coresearch: German grammar as position + morphology evidence — explor…
claude Oct 4, 2026
35383c9
coresearch map: re-file ranks 18 and 19 per the second premise gate
claude Oct 4, 2026
873699b
deepnsm-v2: grammar reads by ReadParams, not thinking styles; WordNet…
claude Oct 4, 2026
45d341d
board: WordNet gives the address, COCA + tokens give the torque (D-LX…
claude Oct 4, 2026
098d49d
knowledge: Bible lanes are not independent witnesses (Aramaic-derived…
claude Oct 4, 2026
eb5aed7
knowledge: why the TEKAMOLO reader is right-corner, even for German
claude Oct 4, 2026
e3ec0f6
deepnsm-v2: first probes from the German grammar map (D-LXC-23)
claude Oct 4, 2026
439b63a
Merge remote-tracking branch 'origin/main' into claude/peaceful-gauss…
claude Oct 4, 2026
915ed7e
deepnsm-v2 ud_case_eval: time vs place reading of Wechsel preposition…
claude Oct 4, 2026
88dbee9
deepnsm-v2: prepositional object vs adverbial; abstract-noun class (D…
claude Oct 4, 2026
04baf7c
deepnsm-v2: TIME / PLACE / FIG lane quorum for Wechsel phrases (D-LXC…
claude Oct 4, 2026
1349006
deepnsm-v2 lane quorum: confirmation round on fresh labels (D-LXC-25)
claude Oct 4, 2026
80760f9
board: restore the D-LXC-25 confirmation-round section cut by the pre…
claude Oct 4, 2026
a584ff3
Merge remote-tracking branch 'origin/main' into claude/peaceful-gauss…
claude Oct 4, 2026
60c2ed1
deepnsm-v2 lane quorum: adverbs of time, including adjectives used as…
claude Oct 4, 2026
a737988
deepnsm-v2 ud_case_eval: gender + number decide the article's case (R…
claude Oct 4, 2026
f2ae2d9
deepnsm-v2 data: record the DeReWo / DeReKo-2014 German frequency lis…
claude Oct 4, 2026
0fa066d
deepnsm-v2: gender of unseen German nouns via compound head, lemma an…
claude Oct 4, 2026
af6a4cd
deepnsm-v2: hydrate joins each marker to its own node; toc_hydrate pa…
claude Oct 4, 2026
b0598a6
deepnsm-v2: DeReKo profile and TEKAMOLO order across genres (D-LXC-28)
claude Oct 4, 2026
efb0fe3
deepnsm-v2 toc_hydrate: measure 6 vs 8 rails per 16-byte tenant on KJ…
claude Oct 4, 2026
e4c4783
board D-LXC-29: group nodes are explicit (AD-like) membership, not st…
claude Oct 4, 2026
0374c45
board D-LXC-29: 16-byte register as 16 x u8 / 32 x u4 tags; clause-le…
claude Oct 4, 2026
9062218
board D-LXC-29: 128-bit classid-free tenants via hot-plug ReadMode / …
claude Oct 4, 2026
d8edb74
deepnsm: CodeRabbit round on #1321
claude Oct 4, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 19 additions & 2 deletions .claude/board/STATUS_BOARD.md
Original file line number Diff line number Diff line change
Expand Up @@ -27,15 +27,32 @@ Plan: `.claude/plans/deepnsm-v2-lexical-evidence-consumer-v1.md`. Convergence br
| D-id | scope | status | gate / falsifier |
|---|---|---|---|
| **D-LXC-1** | `bible_wave` only: counted pick (fold in the example) replaces the word_forms layer's first-wins; lemma table stays first (variant B, pinned rule F9 from `ec50f07b`); library unchanged. Rewritten 2026-09-29 against the migration documents, every claim re-read | Queued | G1–G7 in the plan; G5 KJV before/after blocking; G6 exactly 25 tag changes within `bible_vocab.txt`; G7 restores the deleted lemma-first test (closes `TD-DEEPNSM-V2-SESSION-RESIDUE` item 3) |
| **D-LXC-2** | FSM (`Tagged`/`parse_to_spo`) takes several readings per token, resolved by role (`E-SURFACE-FORM-COLLAPSE-1`) | In PR | `fsm::parse_readings` + `PosSet` + `coca` fold; T1-T7 disable-verified; KJV accounting pinned in `bible_wave` (3,363 ambiguous tokens / 141 words, 1,908 narrowed, 0 unexplained). Entry `2026-10-03-deepnsm-v2-multi-reading-fsm.md` |
| **D-LXC-3** | lemma-table order: keep first (B, 25 in-vocab changes) or counted-first (A, 141; overrides F9) | Shipped | DECISION 2026-10-03: B — the lemma table order is kept (F9 stands); D-LXC-2 runs on it |
| **D-LXC-2** | FSM (`Tagged`/`parse_to_spo`) takes several readings per token, resolved by role (`E-SURFACE-FORM-COLLAPSE-1`) | Shipped (#1314) | `fsm::parse_readings` + `PosSet` + `coca` fold; T1-T7 disable-verified; KJV accounting pinned in `bible_wave` (3,363 ambiguous tokens / 141 words, 1,908 narrowed, 0 unexplained). Entry `2026-10-03-deepnsm-v2-multi-reading-fsm.md` |
| **D-LXC-3** | lemma-table order: keep first (B, 25 in-vocab changes) or counted-first (A, 141; overrides F9) | Shipped | DECISION 2026-10-03: B — the lemma table order is kept (F9 stands); D-LXC-2 runs on it. ⊘ 2026-10-04: the first row no longer settles a noun/verb homograph — D-LXC-13 widens it and position decides |
| **D-LXC-4** | `academic_20k.csv` loader + the duplicate `coca_pos` in `genre_shapes.rs` | Blocked | ruling on 3 duplicate (word, PoS) pairs: wastewater/n, disproportionately/r, instill/v |
| **D-LXC-5** | KJV-own counts for the Cam96 vocabulary (`bible_vocab.txt` is surface-only) | Queued | computed from the corpus |
| **D-LXC-6** | mask-risc fold result at version v into cycle v+1 as a staged cast | Queued | own plan; touches the reverted CE64 area; not re-verified in the 2026-09-29 rewrite |
| **D-LXC-7** | repo `.claude/settings.json` attribution off | Queued | rides with D-LXC-1 |
| **D-LXC-8** | scrub the model-naming `Co-Authored-By` trailer from history (120 commits on `main`) | Deferred | operator: no force push |
| **D-LXC-9** | `archaic_pos` cannot override COCA-known KJV words (`art` is Noun, `word_forms.csv:1047`) | Queued | pre-existing |
| **D-LXC-10** | `Pos::Rel` produced by no in-crate tagger in deepnsm-v2 (only FSM match arms + one unit test) | Queued | pre-existing; external callers can still pass `Pos::Rel` through the public `Tagged`/`parse_to_spo` |
| **D-LXC-13** | noun/verb homograph decided by position: lemma N/V tag widened to its other reading (`coca::predicate_alternatives`), slot rule after a fresh subject | In PR | slot rule + widening disable-verified; KJV re-pinned (34,824 ambiguous / 451 words, 0 unexplained). Current pin after the floating-quantifier and carried-subject fixes: certain 57,325 + alternative 29,287, reading accounting 14,111 / 20,713 / 1,898 / 155→161. Entry `2026-10-04-deepnsm-v2-predicate-by-position.md` |
| **D-LXC-14** | `Pos::Adv`; `Typology` (measured, gates adjective licensing); `AttributeRule` adjective/adverb clauses (off: below frequency); `ud_pos_eval` gold harness (en/de/fr) | In PR | noun/verb with COCA: position 96.7 % vs frequency 85.1 % on 457 decided UD-EWT tokens; adjective/adverb: no clause beats frequency in every language. Entry `2026-10-04-deepnsm-v2-position-rules-on-ud-gold.md` |
| **D-LXC-15** | KJV silver noun/verb labels from Luther 1545 casing (release `v0.1.0-codebooks-2026-07-26`); clause rule as an off-by-default switch; WordNet as a parallel coordinate | In PR | KJV: slot + licensing 85.8 % vs COCA lemma tag 74.9 % on 756 decided; clause rule below slot rule and frequency in en/de/fr/KJV; WordNet chooser 59.3 %, existence filter changes 1 form. Entry `2026-10-04-deepnsm-v2-kjv-silver-labels-from-luther.md` |
| **D-LXC-16** | learned position table (neighbours' readings + copula → P(reading)), German case kept, Animal Farm silver reference | In PR | position × frequency beats frequency on adjective/adverb in 5/5 settings (Animal Farm 78.1 → 88.4 %), noun/verb in 4/5. Harness only. Entry `2026-10-04-deepnsm-v2-learned-position-table.md` |
| **D-LXC-17** | German TEKAMOLO cues by verb position (`tekamolo_de`); adverbial adjectives → Adv fold; Te-Ka-Mo-Lo order law measured | In PR | cue words: position × frequency 88.5 % vs frequency 84.1 % (208 tokens, `da` 87.5 vs 37.5); order law 46–65 % on gold, not usable. Entry `2026-10-04-deepnsm-v2-german-tekamolo-verb-position.md` |
| **D-LXC-18** | question test as the Pearl 2³ mask (`fsm::answered_questions`, `UD_CTX`); blast radius of deleted grammar heuristics | In PR | 2³ context: Animal Farm adj/adv 91.8 % on decided (frequency 78.4 %), German adj/adv 85.1 % (best); noun/verb better with neighbours. German TEKAMOLO capability was lost in `68955ecb`; tenant writer reads the condemned v1 WordNet rail. Entry `2026-10-04-deepnsm-v2-question-test-2cube-and-blast-radius.md` |
| **D-LXC-19** | literal rules, priorities and a quorum for word-class homographs (`ud_pos_eval` `UD_RULES`, `UD_DE_INVENTORY`) | In PR | 14–20 rules per pair; held-out-fitted joint quorum best or tied in 7/10 rows (UD English n/v 93.7 %, de n/v 98.8 %); summed quorum fails on correlated voters (de n/v 84.9 %); Animal Farm adj/adv and German adj/adv stay best with position × frequency. Entry `2026-10-04-deepnsm-v2-rule-quorum.md` |
| **D-LXC-20** | Frageprobe through the Satzklammer; `answered_questions` returns `causal_edge::CausalMask` | In PR | German predicative at the copula clause's right edge 88.5 %, manner adverb before the right bracket 90.0 % (vs after-copula 35.3 %); German adj/adv joint quorum 81.6 → 85.1 % (disable: 81.9 %). Entry `2026-10-04-deepnsm-v2-satzklammer-frageprobe.md` |
| **D-LXC-21** | deleted grammar heuristics restored (deepnsm-v2 toc/hydrate/promote/lexicon/loci/tekamolo + toc_hydrate; deepnsm ontology_vocab + persona_chain_replay); WordNet identity register + ConceptMask | In PR | toc_hydrate on KJV: 32,357 HHTL nodes, 70,393 triples addressed, 1,227 basins at TOC keys; ContextChain margin 335× too coarse for the rank metric. Ledger `.claude/knowledge/deleted-grammar-heuristics-ledger.md`; entry `2026-10-04-deepnsm-deleted-grammar-restored.md` |
| **D-LXC-22** | DECISION: WordNet hypernym metric may feed ndarray CLAM/CHAODA offline as an evaluation instrument (identities unchanged, `48405aa2` stands); grammar is not a thinking style → `tekamolo::read_clause` takes `ReadParams`, style adapter kept for the planner / ogar-loco | In PR | presets pinned to the adapter rows (disable-verified); 195 lib tests. Entry `2026-10-04-wordnet-clam-chaoda-scope-and-grammar-read-params.md` |
| **D-LXC-23** | first probes from the German grammar map: der/die/das DET vs PRON voters (`ud_pos_eval`), `ud_case_eval` (fixed-case prepositions, article decidability, Wechselpräposition, relative pronoun), `tekamolo_corners` | In PR | DET/PRON German 92.3 → 96.0 % (pronoun P 90.0 / R 54.1) PASS; case on HDT: Acc 90.0 / Dat 92.5 % PASS, Gen KILL, Wechsel article-decided 99.7 %, verb prior KILL (63.0 vs 63.9), relative pronoun 80.1 vs 63.8 % PASS; right-corner-on-Bible prediction KILL (Luther 7.1 % vs GSD 12.3 %). Entry `2026-10-04-deepnsm-v2-first-probes-from-the-grammar-map.md` |
| **D-LXC-24** | prepositional object vs adverbial (`ud_pp_arg_eval`): the (verb, preposition) object rate takes verb-governed phrases out of the TEKAMOLO pool and fixes their case; abstract-noun suffix class | In PR | HDT (gold = `obj` + ADP; German `obl:arg` is bare datives): A1 P 77.2 / R 81.6 %, 896 of 1,098 objects removed, 264 of 19,466 adverbials lost, PASS; A2 case 96.8 vs 81.0 % PASS; abstract heads 10.7 % objects vs 4.6 %; R3a KILL on HDT. Entry `2026-10-04-deepnsm-v2-prepositional-object-detector.md` |
| **D-LXC-25** | TIME / PLACE / FIG lane of a Wechsel phrase (`ud_lane_quorum`): noun, number, adverb, verb governance, TEKAMOLO position and article voters, combined by logistic regression; silver labels | In PR | abstract head never PLACE: PASS, replicated (3.8 %, 3.2 %). Confirmation round (fresh labels, frozen design): noun voter TIME F1 0.831 PASS, PLACE F1 0.476 KILL, quorum +0.5 over noun KILL. Entry `2026-10-04-deepnsm-v2-wechsel-lane-quorum.md` |
| **D-LXC-26** | gender + number of the head noun decide the article's case (`ud_case_eval` R6); stemmer fallback via `frostem` German Snowball (the tesseract-paperless search stemmer) | In PR | train case purity 65 → 79 % with gender + number; HDT R6 76.4 vs 64.1 % on seen nouns (+12.3); stem fallback 63.7 % KILL (the stem erases number); exploratory gender-only fallback 68.4 vs 66.9 %. Entry `2026-10-04-deepnsm-v2-gender-number-case.md` |
| **D-LXC-27** | gender of unseen German nouns (`ud_gender_eval`): compound head, DeReKo lemma, `frostem` stem, nominalised infinitive | In PR | HDT unseen nouns: compound head covers 83.7 % at 98.1 % PASS; stem 88.0 % PASS; lemma 98.4 % on 4.7 %; infinitive 91.5 % KILL; cascade 89.9 % coverage at 97.7 %. GSD weaker (compound 88.8 % KILL). Entry `2026-10-04-deepnsm-v2-unseen-noun-gender.md` |
| **D-LXC-28** | DeReKo-2014 profile across genres (`dereko_profile`); TEKAMOLO order by position in edited novels vs Bible translations (`tekamolo_order`) | In PR | ADJD about 3.4 % of tokens in novels vs about 1 % in every Bible; order: fresh novels 58.9 % vs Bibles 47.1 %, intervals overlap, KILL (z ≈ 2.1); density Te/Mo 3–5× lower in Bibles (robust). Entry `2026-10-04-deepnsm-v2-dereko-and-tekamolo-order-across-genres.md` |
| **D-LXC-29** | 6 vs 8 rails of u8:u8 per 16-byte tenant (classid already in the row key), measured on KJV basin membership (`toc_hydrate` TENANT CAPACITY) | In PR | basins per verse median 3 / p95 6 / max 12; one tenant fits 95.66 % (6 rails) vs 99.41 % (8 rails); overflow 1,106 → 149; exact-set group nodes do not deduplicate (1,099 of 1,106 unique). Layout change OPEN: needs an envelope audit and an operator ruling. Entry `2026-10-04-deepnsm-v2-tenant-rails-6-vs-8.md` |

## D-RPT — ReportPlan / zero-copy pivot / report-as-OGAR-projection (2026-09-23)

Expand Down
8 changes: 4 additions & 4 deletions .claude/board/SUPERSESSION-INDEX.md
Original file line number Diff line number Diff line change
Expand Up @@ -62,7 +62,7 @@ coverage instead.
| `ResonanceDto` | REPURPOSE | `PerturbationDto` | 2 | 12 | 7 |
| `BindSpace` | RETIRE | — | 68 | 48 | 42 |
| `CollapseGateEmission` | RETIRE | — | 5 | 14 | 12 |
| `ThinkingStyle` | RETIRE-toward-contract | — | 52 | 28 | 24 |
| `ThinkingStyle` | RETIRE-toward-contract | — | 53 | 28 | 24 |

## Table 2 — plans naming a ruled symbol without citing the ruling (76)

Expand All @@ -83,7 +83,7 @@ a licence to act on it.
| **READ** | `archetype-scaffold-v1` | `GateDecision` | In progress (2026-04-24) | 0/0 |
| **READ** | `capstone-cognitive-loop-wiring-nan-census-v1` | `GateDecision` | PROPOSED (2026-06-20). The measurement compa | 0/0 |
| **READ** | `deepnsm-v3-convergence-v1` | `StepMask` | PROPOSED (doc-only). Extends `v3-convergence | 0/5 |
| **READ** | `epistemic-quadrant-materialization-v1` | `MergeMode` | PROPOSED.** Operator direction 2026-07-29: * | 1/4 |
| **READ** | `epistemic-quadrant-materialization-v1` | `MergeMode` | PROPOSED.** Operator direction 2026-07-29: * | 2/4 |
| **READ** | `graphrag-doc-retrieval-soa-integration-v1` | `GateDecision` | DESIGN + FIRST CODE. **v1.2 (2026-07-17):** | 3/10 |
| **READ** | `hexagon-plasticity-v1` | `MergeMode` | ACTIVE — landed 2026-09-14 on operator go. * | 2/10 |
| **READ** | `integration-actionhandler-rbac-orchestration-v1` | `GateDecision` | HARDENING (5+3 in progress). | 0/0 |
Expand All @@ -96,8 +96,8 @@ a licence to act on it.
| **READ** | `persistence-artifact-backed-commit-v1` | `persist_cycle` | RATIFIED (operator ruling 2026-08-09). Phase | 0/0 |
| **READ** | `post-teardown-buildup-survey-v1` | `StepMask` | SURVEY, read-only, plan-only (no code, no te | 1/5 |
| **READ** | `r2il-bpe-typed-genetic-recombination-v1` | `GateDecision` | PROPOSAL, §7's three falsifiers now RUN (see | 1/1 |
| **READ** | `scientific-kg-substrate-v1` | `GateDecision` | PROPOSED — **scoping doc**, no code. Records | 0/9 |
| **READ** | `self-reasoning-substrate-v1` | `GateDecision` | PROPOSED — doc-only. No code, no contract ch | 0/5 |
| **READ** | `scientific-kg-substrate-v1` | `GateDecision` | PROPOSED — **scoping doc**, no code. Records | 3/9 |
| **READ** | `self-reasoning-substrate-v1` | `GateDecision` | PROPOSED — doc-only. No code, no contract ch | 4/5 |
| **READ** | `v3-convergence-wiring-v1` | `GateDecision` | ACTIVE (2026-07-01). Operator: "I'm all in f | 0/0 |
| **RESCOPE** | `soa-migration-diff-resolution-2026-06-13` | `BindSpace`, `CollapseGateEmission`, `GateDecision`, `MergeMode` … | — | 1/5 |
| **RESCOPE** | `cognitive-substrate-convergence-v1` | `BindSpace`, `CollapseGateEmission`, `GateDecision`, `MergeMode` … | PROPOSAL (sprint-10 architectural decisions | 3/13 |
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,49 @@
# 2026-10-04 — /coresearch: German grammar as position + morphology evidence (exploration map)

**Status:** OPEN — an exploration map. Nothing is adopted; the operator chooses.
Map: `.claude/knowledge/german-grammar-rule-inventory.md`.

**Council:**
- Brief and premise gate: PREMISE-SPLIT into Q1 word class, Q2 case, Q3 anaphora, Q4 TEKAMOLO, Q5 plausibility.
- 5 scouts: code, internal prior art, literature, linguistic resources, grammar concepts.
- A 23-row crosswalk.
- 3 co-architects: bridge, firewall, falsifier.

## Key results

**Already have.**
- Satzklammer adj/adv rules: 88.5 % / 90.0 %.
- German adj/adv quorum: 85.1 %.
- The TEKAMOLO order law is killed as a decider; the cue-by-verb-position rule scores 88.5 %.

**Inventories without consumers.**
- Five mined TSVs have no Rust reader: `article_decidability`, `valency`, `wechsel`, `relative_pronoun`, `reflexive`.
- They become one Q2 case-constraint evaluator (`ud_case_eval`), not five voters.
- Shared precondition: `next_form` in `Tok`, plus the tables re-mined on the fit split.

**Data facts that change the design.**
- UD German has `iobj` = 0: a dative object is obj/obl + Case=Dat.
- ADP carries no Case.
- GSD Case gold is about 10 % noisy, so Case kills must hold on HDT.

**Cheapest probes first.**
- Fixed-case prepositions.
- der/die/das DET vs PRON.
- Relative-pronoun case.
- Paradigm-exact monoflexion (read on HDT).
- Expletive es.
- Separable particles.
- Auxiliary selection.

**Operator additions (2026-10-04).**
- **Belief transitions:** re-filed by the second premise gate. Revision history is a planner belief-arena record (magnitudes ordered across books), not 24×i4 loci; only a peer pointer within ±8 may be a locus. Probe: the history must predict later contradiction beyond the final TruthValue.
- **WordNet → ndarray CLAM/CHAODA:** outside the letter of `48405aa2`, inside its spirit. Clean only offline, under a recorded DECISION (draft wording in the map, §6). Second premise gate: a path distance is symmetric, so S↔O swaps are invisible; points must be role-indexed `(verb, role, synset)`, or the items restricted to verb swaps.
- **hydrate = node creation:** ALREADY-HAVE (restored, green on the KJV).

## Not searched

- adverbial vs predicative ADJD accuracy;
- a computational Wechselpräposition accuracy;
- German Winograd beyond Wino-X;
- adverbial-order corpus studies;
- left-corner parsing for German.
Original file line number Diff line number Diff line change
@@ -0,0 +1,53 @@
# 2026-10-04 — deleted grammar heuristics restored; WordNet identity register (D-LXC-21)

**Status:** VERIFIED-IN-CODE and MEASURED (`cargo test --release` deepnsm-v2: 194 lib tests; deepnsm: 222; `toc_hydrate` and `persona_chain_replay` run on real data, debug-0).

Operator (2026-10-04): restore the deleted grammar heuristics, repurpose or merge the superseded ones, drop none, and write down which commit had what. Ledger: `.claude/knowledge/deleted-grammar-heuristics-ledger.md`.

## Restored

**From `68955ecb`, into deepnsm-v2:**
- `toc.rs`, `hydrate.rs` and `promote.rs`: the KJV TOC as an HHTL tree, triples addressed to verse nodes, and basins promoted at TOC keys.
- `lexicon.rs`.
- `loci.rs`.
- `tekamolo.rs`: the German right-corner lanes.
- the `toc_hydrate` example.

**Into deepnsm:**
- `ontology_vocab.rs`, from `48405aa2`.
- the `persona_chain_replay` example, from `8c03aaa9`.

## Repurposed

**`ontology_vocab.rs` now holds a WordNet identity register:**
- `WordNetRail` reads every sense and every hypernym edge.
- `ConceptMask` covers a sense when the sense or any ancestor is a mask root, so a mask can cut across the taxonomy (e.g. water animals).
- There are no distances, so A1 holds.
- Its OBO half answers empty, because `98de36eb` retracted the 0x03 rows; a test pins that.
- Disable runs:
- is_a without the ancestor walk → 3 of 5 tests red;
- every sense collapsed to number 1 → 2 red.

## Measured

**`toc_hydrate` on the KJV:**
- 32,357 nodes (66 books, 31,102 verses);
- 70,393 triples, 0 unaddressed;
- 38.8 % cross-verse links;
- 1,227 basins at TOC keys;
- loci: Recency binds 35,025 of 35,613 pronoun subjects, SelectionalFit binds 27,788.
Comment thread
coderabbitai[bot] marked this conversation as resolved.

**`persona_chain_replay`:** the ContextChain margin gate is 335× too coarse for the rank metric (spread 0.0003 vs threshold 0.1), so every replay escalates.

## OPEN

- `tekamolo.rs` still carries `V2StyleProvider`. §3o refused thinking styles in deepnsm-v2; the operator decides whether to keep it here or move it to the planner.
- `loci.rs` is agreement-blind. The shipped agreement resolver scores 0.727 against 0.273 for the agreement-blind rule. Merge queue in the ledger.

> **⊘ Updated 2026-10-04 (PR #1321 review, same day).**
> - `toc_hydrate` now parses with every reading (`parse_readings(...).certain`, the same stream as `bible_wave`).
> - The FSM no longer treats an object carried into the subject slot as a subject head.
> - Current KJV numbers: **57,325 triples** (= `bible_wave`'s `certain` pin), 0 unaddressed, 1,187 basins.
> - `persona_chain_replay` candidates are now nouns only (the resolver's own lexicon): 2 replays run, both escalate, and the margin gate is **617×** too coarse (spread 0.000162 vs 0.1).
> - deepnsm-v2 lib tests: 196.
> - The figures above are the first run's.
Loading
Loading