docs(tri-comparison): mandatory recall audit — every GG non-detection assessed & logged

[squid-protocol]

What

Adds step 2.6 to the tri-comparison-ledger-sweep skill — a validation gate requiring
that every function tree-sitter OR ctags reports that GitGalaxy does not, across the whole
corpus, is individually read and sorted into exactly one of two buckets:

1. a real GitGalaxy recall gap → GitHub issue with an isolated repro
2. a comparison/audit-tool artifact → name the mechanism (macro hallucination, #if 0 dead
   code, bodyless declaration, deliberate scope boundary, naming-convention mismatch,
   occurrence-alignment fuzz), fix the audit or document it

Why

The ledger keys entries by discrepancy pattern. A shape verdict of "tree-sitter over-detects
here" can be marked validated without anyone confirming every occurrence in that bucket is a
tree-sitter error rather than a GitGalaxy miss. Confirmed real (cpp, 2026-08-29): one bucket
held ~164 occurrences — ~162 tree-sitter OPCODE-macro / _FORCE_INLINE_-parse artifacts, but
2 genuine GitGalaxy recall gaps (STDAPI DllCanUnloadNow() macro-return-type form; K&R
PRIVATE void PlinkPrint(out,plp,tag) untyped-param defs) that the blanket "over-detects"
verdict would have buried forever.

Changes

• SKILL.md step 2.6 — the gate, how to run it, the two buckets with per-mechanism
  resolution guidance.
• Step 8 capstone trigger — now requires the recall audit complete, not just a fully-
  validated ledger (a validated ledger with an unaudited recall gap is exactly the failure
  mode this prevents).
• Capstone §9 must carry: the recall number, "every non-detection individually assessed",
  the bucket-1 count (each linked to its issue), and the bucket-2 per-mechanism breakdown with a
  real file:line each.
• tests/tools/recall_audit.py (new) — enumerates every non-detection vs tree-sitter and
  ctags at the occurrence level, with source lines, and cross-checks measure()'s own
  real_functions - found_functions so a divergence between the raw name-diff and the counted
  misses is itself surfaced.
• tri_comparison_README.md — a "is GitGalaxy missing anything?" section with a
  RECALL_AUDIT marker block for the standing answer.

Follow-up (separate PR)

Executing step 2.6 for the current state: the 8 languages with func recall < 100% break down as
~3 genuine GitGalaxy gaps (shell keyword-as-argument, c K&R defs, cpp macro-return-type) + a
large tail of tree-sitter/ctags artifacts the audit is currently counting against GitGalaxy's
recall. That PR files the issues and corrects the audit.

🤖 Generated with Claude Code

https://claude.ai/code/session_019Zm1uVBpVEZJ9SB7bWFR8S

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab

[squid-protocol]

Superseded by #2463, which includes this commit plus the execution of step 2.6 (c/cpp/fortran/objc recall corrections + docs). Closing in favor of that one.