First vault-lint — 104 raw findings, zero truth-defects, and what that means for the consolidation question

First lint in the vault's history, run at 63 notes in 30-notes/ (threshold 50, operating spec §6). Mechanical pass = 00-meta/tools/vault_lint.py (the five authorized checks, scripted, no editorializing). This note is the judgment layer. Lane rule in force tonight (2026-07-07, model constraint): judgment commentary below is written only for in-lane notes; every out-of-lane note is listed at the bottom with the structural-check line only. Deferred ≠ rejected.

Scripted results (verbatim from the tool)

• scanned 84 files, 435 wikilinks
• frontmatter: 4 · audit_status missing: 0 · dead links: 81 · orphans: 0 · word count >400: 19 → TOTAL 104
• supplementary spec-§6 counts (same run, same script discipline): stale seedlings >30 days: 15 · inbox unpromoted >14 days: 0
• not run tonight, named so nothing silently vanishes: live URL link-rot sweep (network-wide check is not one of the five authorized checks); contradiction detection (no mechanical detector exists — the one known flagged contradiction, special-case vs precursor on Kelley-Bryson, is already on record from cycle 8)

What the 104 decompose into (judgment, in-lane)

Dead links (81). Not eighty-one problems — three:

1. 74 are concept-style red-links (training, KV cache, syllogism…) confined almost entirely to the twelve inference-cluster notes dated 2026-06-04, plus three in two Linnainmaa notes. This is one era's writing style, not link rot: those notes were written Wikipedia-style, red-linking every term as a potential future note. Nothing since June 4 does this. One ruling ("keep as future-note markers" vs "strip to plain text") closes all 74 at once — queued for Cali, not decided here.
2. 6 are claim-inference-engineering-emerged-as-specialty — the dangling link cycle 4 explicitly ruled leave standing (reasons in the retired bridge question). Already-ruled; not re-litigated.
3. 1 false positive — question-gap-detection.md uses [[wikilink]] as a word in prose. Tool limitation, noted for the next lint.

Frontmatter (4) and the long-form word counts (4 of the 19). The same four notes both times: the May-era long-forms (backpropagation-gap, introspection-access, cot-faithfulness, rationality-continuity), all pre-dating the claim-note schema, all already carrying honest audit_status markers (pre-schema-era or flagged (V-0NN)). The lint re-found what the audit already found. No new information; grandfathered.

Word counts (the other 15). Claim-notes at 403–857 words. Two causes, both structural: (a) the June-4 inference cluster again (era style), and (b) the Linnainmaa cluster, where append-only revisit-history sections — which the vault is right to keep — push otherwise-compliant notes past the 400-word ceiling. The revisit discipline and the word ceiling are in mild tension; that tension is consolidation-question evidence, not a defect in any single note.

Stale seedlings (15). Every note from the May/June era still says status: seedling — including notes that have since been revisited, re-sourced, and MOC-linked. The status field simply never advances. Hooks don't cover status hygiene; lint does. This is the clearest example of a lint-only catch.

One missing MOC (in-lane, lane 6). The gap-detection cluster is now six notes (bridge-detection, GRADE, kb-completeness, llm-explicit-implicit, obligatory-attributes, query-failure-clustering) with no MOC — spec threshold is 5+. Queued for the next cycle's CONNECT step, where fixes belong.

The catch-count, as the sequencing ruling defined it

"Count what lint catches that no forward-hook had flagged": ~31 hygiene items, 0 truth-defects. The 31 = 15 stale statuses + 15 word-count overages + 1 missing MOC. The 0 is the number that matters: no orphans, no missing audit_status, no duplicate claims, no contradiction surfaced that wasn't already flagged, no stale value served as current — i.e., none of the reconciliation-class failures that Dreams-style global consolidation exists to fix. Everything truth-critical had already been caught by the targeted mechanisms (audit flags, forward-hooks, myth status histories) before lint ever ran. Full reading appended to question-consolidation-pass-vs-revisit-protocol.

Out-of-lane notes (structural check only, per tonight's lane rule)

Each of the following: structurally checked, commentary deferred-model-constraint 2026-07-07.

• claim-cognitive-conflict-prerequisite-for-restructuring
• claim-expedient-knowledge-blocks-restructuring
• rationality-continuity-problem
• introspection-access-problem
• cot-faithfulness-anthropic-biology
• claim-selfaware-canonical-self-knowledge-benchmark
• claim-llm-neural-metacognition-incomplete
• moc-machine-self-knowledge

(For the four of these that appear in the scripted findings above, the scripted lines stand as reported — scripts don't editorialize — and the judgment-level reading resumes on the next Opus session.)

Pattern worth keeping

The lint's findings are almost perfectly stratified by era: pre-schema May long-forms, Wikipedia-style June 4, schema-era July. Nothing written under the current schema tripped anything except word counts inflated by honest revisit appends. The vault's problems are fossils, not leaks — the discipline installed by the audit is holding at the write boundary, and the lint's real job turned out to be archaeology, not firefighting.

— Seek (first vault-lint, run by the queen on the 2026-07-07 Fable window)