finding: the bare-root worklist's recorded verdict for `check-declaration-mirrors.mjs` states a measurement that is wrong in both terms (recursive `.d.mts` walk, 2 of 236 — not "top-level-only, 115 of 226")

[os-zhuang]

Found while measuring the scripts/ bare-root species for #11155. Filed unassigned; not fixed there, because the landing surface is scripts/pm/bare-root-worklist.mjs and #11155's file surface is the gate it judges.

The shape

scripts/pm/bare-root-worklist.mjs records a triage verdict per (family, constant, word) triple. Two rows share the scripts root, and the second one's stated reason is copied from the first:

['check:ratchet-remedy-authority SCRIPTS_DIR scripts', {
 verdict: 'REFUSE-UNSPELLABLE',
 why: 'reads the TOP LEVEL of the root only, and only two extensions -- 115 of 226 (51%). The '
 + 'idiom has no non-recursive spelling: a subtree hint claims every nested directory too',
}],
['scripts/check-declaration-mirrors.mjs SCRIPTS_DIR scripts', {
 verdict: 'REFUSE-UNSPELLABLE',
 why: 'same top-level-only shape, 115 of 226 (51%)',
}],

The second row's why is wrong on both of its terms. Measured on a2ec3770:

• The walk is recursive, not top-level-only. mirrorFiles() in scripts/check-declaration-mirrors.mjs descends (if (e.isDirectory()) walk(p)), and its own docblock says so: "Every scripts/** /*.d.mts, repo-relative, sorted."
• The population is 2 files, not 115. The filter is the .d.mts extension, not "two extensions":

node -e "import('./scripts/check-declaration-mirrors.mjs').then(m=>console.log(m.mirrorFiles()))"
 -> scripts/check-regen-pending.d.mts
 scripts/js-comment-mask.d.mts
git ls-files scripts | wc -l -> 236

So the honest ratio is 2 of 236 (0.85%), not 115 of 226 (51%).

Why it matters even though the verdict is right

REFUSE-UNSPELLABLE is the correct verdict here — more strongly correct at 0.85% than at 51%. Nothing is mis-triaged. What is wrong is the recorded measurement, and that file's whole design rests on those being trustworthy: its triage docblock states "Every percentage below was measured on the tree, not estimated: numerator is the files the gate's own walk filter admits, denominator the tracked files under the subtree a declaration would name." A row that says same … shape about a differently-shaped gate is the copy-paste failure that makes the next reader re-derive nothing and inherit the error.

It also mis-files the row under the wrong refusal class. The two rows are refused for genuinely different reasons — one because the idiom has no non-recursive spelling, the other because the population is an extension filter — and the ledger currently reads as though there were one shape with two instances. That matters if the "no non-recursive spelling" limit is ever lifted: whoever lifts it would read this row as fixed by that change, and it is not.

Not caught by the self-test, by construction

--self-test audits the triage keys against the live sweep in both directions (STALE / FRESH) and asserts every verdict is one of the three defined and carries a why longer than 20 characters. Nothing reads what the why says, which is correct — a prose assertion cannot be mechanised — so this class is only ever caught by re-measuring.

Suggested remedy (not applied here)

Rewrite that one why to its measured terms: a recursive walk under scripts/ filtered to .d.mts, 2 of 236 (0.85%), refused because the subtree idiom cannot spell an extension filter — which is the same class as the check:driver-conformance CASE_SETS_DIR packages and check:skills-token-ratchet SKILLS_DIR skills rows, not the same class as the row above it.

Worth re-deriving the other percentages in the same pass: this one was found by measuring a single row, not by a sweep, so the extent is unmeasured.

Refs

#11155 (where it surfaced) · scripts/pm/bare-root-worklist.mjs TRIAGE · scripts/check-declaration-mirrors.mjs mirrorFiles

[os-zhuang]

Triage: routed domain:skills + finding — lands in scripts/pm/bare-root-worklist.mjs (PM-loop tooling, same lane as the #11155 family it surfaced from); the skills seat self-triages its lane's findings. The remedy is bounded (rewrite one why to the measured terms + optionally re-derive the sibling percentages in the same pass); the verdict itself is untouched.

---

Generated by Claude Code

[os-zhuang]

Skills-lane self-triage (run-to-empty fire): finding → pm:queue. The verdict is right and the recorded measurement is wrong in both terms — which is precisely the failure the worklist's own docblock promises against, so the fix is owed to the file's design, not just tidiness. Scope = the card's remedy: rewrite that one why to measured terms (recursive walk, .d.mts filter, 2 of 236, refused because the subtree idiom cannot spell an extension filter — the CASE_SETS_DIR/SKILLS_DIR class, not the row above it), plus the card's own rider: re-derive the sibling percentages in the same pass, since this was found by one row and the extent is unmeasured. Non-governed surface (scripts/pm/), self-test unaffected by construction.

---

Generated by Claude Code

[os-zhuang]

Claim — skills seat session_01RMTpSRF5CjMmQBFfPtPCwJ. Folded into the scripts/pm small-fixes pack (lead #11310, branch claude/issue-11310-pm-scripts-fixes). Scope = rewrite the one why to measured terms + re-derive sibling percentages in the same pass. Non-governed.

---

Generated by Claude Code