anneal

/anneal

Intent. A patient pass that lowers the disorder of the graph without adding new content. Reviews recent work against the methodology core; flags drift; offers fixes; applies on user confirmation.

The metallurgical metaphor is the right one: you are not creating; you are bringing existing structure to a more ordered, lower-energy state. Slow heat, careful cooling. Never proactive — only invoked.

Scopes

Scope	What's in scope
--scope=note <path>	one note
--scope=topic <topic-slug>	one topic-MOC + every claim it owns
--scope=vault	the whole vault (sampling for very large vaults)

If no --scope, default to "all notes modified in the last 7 days".

Behaviour

Phase 1 — gather

1. Resolve scope into a concrete list of note paths.
2. Read each note: title, frontmatter, body.

Phase 2 — assess

For each note, check against the methodology core:

• Title is a prose proposition. Not a topic label. (reference/empirical-grounding/title as claim enables traversal as reasoning.md)
• Description adds information. Not a paraphrase of the title. Layers heuristic → mechanism → implication. (reference/empirical-grounding/good descriptions layer heuristic then mechanism then implication.md)
• Frontmatter contract complete. All required fields per type. Enums valid. Source link present for claims.
• Body length 150–400 words (claims/memories). Outside the band → flag for split or expand.
• Inline wikilinks have surrounding prose. Bare [[a]] [[b]] = drift.
• MOC membership. Every claim/memory should appear in at least one _<topic>.md. (reference/empirical-grounding/MOCs are attention management devices not just organizational tools.md)
• No dangling links. Every [[wikilink]] resolves to an existing file.

For --scope=topic add MOC-specific checks:

• The MOC's ## Core claims lists ≥3 claims (else: too narrow, consider absorbing into parent).
• MOC density (paired-axis check, replacing the old single-axis "≤30 claims" rule). Run walk/tools/topology.py moc-density --threshold N to compute claim count and internal-edge density (mean internal degree among the MOC's members). The verdict combines both:
  • split-candidate — count > threshold AND mean internal degree < 1.0. The MOC is a "cocktail party": many claims that don't talk to each other. Splitting harms nothing.
  • productive-crowd — count > threshold AND mean internal degree ≥ 1.0. The MOC is a "working group" or programme-organising thesis: claims cross-link, density is the evidence engine. Do not suggest splitting. Note the high count for transparency, recommend audit instead.
  • ok — count ≤ threshold.
  • small — count < 3 (consider absorbing into parent).
  • The mean-internal-degree threshold (1.0) is a starting calibration; tune against the vault's distribution before tightening.
• Isolated members (optional extension). topology.py moc-density also lists, per MOC, the claims that have zero links to siblings in the same MOC. These are split candidates regardless of MOC-level density — they're sitting in a tag that doesn't reflect their epistemic position. Suggest re-tagging them or moving them to a more appropriate MOC.
• Bidirectionality: every claim in ## Core claims lists this MOC in its ## Topics footer, and vice versa.