tooling(qa): roll-up view over qa-run records — selector × latest verdict × staleness (#9616)

Answers "what is the latest verdict for every checklist selector, and is it
stale?" in one command, replacing a 23-issue manual read.

A generated VIEW, not a tracker: no cache, no snapshot, no last-run file. The
qa-run issues stay the only source of truth.

Three things were measured over the live corpus rather than assumed, and each
changed the design:

- The title convention has drifted. Neither the documented form
  (.claude/skills/checklist-test/SKILL.md) nor the form described in the card
  is what any record writes: 23/23 use five ·-separated fields with the
  judged/total folded into a parenthetical on the selector, in five distinct
  phrasings, absent entirely on one record. The parser accepts the shape the
  records actually have and PRINTS anything it cannot parse.

- The bodies are not one shape and are deliberately not parsed. 9 of 23 records
  carry no markdown table at all; the other 14 carry 26 distinct header shapes,
  and most of those tables are clause-level rather than item-level — they use
  the word "verdict" for a different unit. Counting them as items would produce
  a confidently wrong matrix. The view says, every run, that it did not read
  them.

- Staleness is three-valued because a shallow clone cannot always tell.
  merge-base --is-ancestor answers false for commits behind the graft boundary,
  so a boolean would have rendered the nine stalest records as fresh. "Could
  not tell" renders as ? and is never counted as fresh.

Reads REST core only (1 call for the whole corpus, 15000/hr quota) — not the
search API and not GraphQL.

--self-test drives the pure core over fixtures taken from the real corpus:
60 assertions, no network.


Claude-Session: https://claude.ai/code/session_01XqDQYVU5smx29ts9pAErja

Co-authored-by: Claude <noreply@anthropic.com>

Author: claude[bot]