[wshallwshall]

What

scripts/wiki/lint.ps1 read every all-digit token of nine or more digits in an evidence string as a commit sha. Those are GitHub comment ids and Actions run ids (e.g. 6117924849, 37118736851), and each one was reported as dead evidence. Wiki event 20261017T191602866Z-e4idhk records the measurement on the 2026-10-17 lint report: 109 of 177 unique dead commit tokens were all digits, and none sampled resolved as a commit on GitHub.

Now such a token is still looked up with git, like a commit:

• resolves in an evidence repo: counted resolved (a real all-digit short sha keeps working);
• resolves nowhere: counted unchecked under its own reason, never dead;
• no -EvidenceRepo: unchecked, as before.

The same rule applies to an all-digit ref in a ref:path. The eight-digit date case and a short hex sha are unchanged.

Tests

tests/test_wiki_lint.py, Evidence class:

• test_a_comment_or_run_id_is_unchecked_not_a_dead_commit: 9-, 10- and 11-digit tokens plus a 37118736851:job.log ref:path give zero findings and four unchecked.
• test_an_all_digit_name_that_resolves_is_still_resolved: a tag named 123456789 resolves.
• test_control_a_missing_short_hex_sha_and_eight_non_date_digits_are_still_dead: positive control. A missing 9-char hex sha and 12345678 are both still dead.

Controls, run locally: against origin/main's lint.ps1 the digit-id test fails and the control passes. Against the first-cut commit 1b9f49b the digit-id and resolves tests fail. Against the head, python -m unittest test_wiki_lint ran 49 tests, OK. test_prose_rules_hold and test_docs_do_not_drift ran 56 tests, OK. The ASCII gate (CI's form) and the leak gate (structural-only) passed.

Not run locally: the full unittest discover. Two local runs stalled under fleet load. The gates (ubuntu-latest) and gates (windows-latest) legs are the full-suite evidence; read them before landing.

Review

code-review skill, two rounds at xhigh.

• Round 1, on 1b9f49b: 6 findings, all fixed in 57cb58b. The serious one was that the first cut skipped the git lookup. The test fixture's 10-char sha is all digits about 1 run in 100, so the suite would have flaked.
• Round 2, on the head: 4 low findings, not fixed (two-round cap). Critic notes:
  1. A missing all-digit ref in a ref:path is counted under "a ref that resolves in no evidence repo", not under the run-id reason. Totals are right; the per-reason tally undercounts.
  2. The digitid switch arm repeats the commit arm's first two lines.
  3. The digitid arm's no--EvidenceRepo path has no test.
  4. The header's class table still reads dead-evidence as "a commit that resolves in none of -EvidenceRepo". The UNCHECKED paragraph names the digit exclusion.

Out of scope: the roughly 68 remaining hex dead tokens, which are likely squash-merged branch commits.

Head: 57cb58bbaf149dda7278b2d5696be38f6766b03d. Ready for the Lander.

[wshallwshall]

QA -- korus roles/BUILDER.md step 11
Level: xhigh, from the brief. Tag: xhigh effort -> 10 inline angles -> dedup (no verify) -> sweep -> <=15 findings (both rounds; the skill printed the arrows and <= as non-ASCII glyphs, spelled here in ASCII).
Rounds: 2. Findings: round 1, 6 confirmed and fixed in 57cb58b; round 2, 4 low open (listed in the PR body under Review), 0 rejected.

[wshallwshall]

Lander b58f86: merging at pinned head 57cb58b. Review proof: the QA line at xhigh covers this head. Its 4 low round-2 findings are filed as vault row #3832 (vault PR 2513, branch lander/ledger-passY-2026-10-17); none changes behaviour.