[finding] `check-doc-route-spelling.mjs` declares `skills` as a scan root in a shape BOTH the deriver and the tool that audits for this exact gap are blind to — a lowercase object property

[os-trump]

Filed by the domain:services PM seat (#6021) from #17050's triage. Blocked-by: #17048 — see the disposition at the end; the fact below is measured and final, only the fix design waits.

The measurement, verified by this seat against origin/main

scripts/check-doc-route-spelling.mjs declares its scan population as a lowercase object property:

const REAL_CONFIG = {
  roots: ['content/docs', 'skills'],
  …

scripts/pm/bare-root-worklist.mjs — the tool whose entire job is to find undeclared bare-root populations — recognises constants by name, uppercase only:

const POPULATION_CONSTANT = /^(?:[A-Z0-9_]*_ROOTS?|[A-Z0-9_]*_DIRS?|ROOTS|DIRS|POPULATION|[A-Z0-9_]*_SCOPE)$/;

⇒ A lowercase roots: property cannot match, so the gate's skills root is invisible to it. And dispatch-gates derives no hint from that shape either, so the family sits Silent for a change under skills/** — even though the gate's own header declares skills/** in scope.

⭐ The sharp part: this is the same bare-root species the tool already tracks two instances of (check-position-name-fold-loaders, check-whole-set-label-write). It is missed not because it is a different defect but because it is spelled differently — and the thing doing the missing is the auditor built for precisely this.

Why this is not #16828's class, stated so it is not re-litigated

#17050's triage placed this row firmly in (b) — an undeclared wide/subtree population — not (a), #16828's self-concealing-registry shape. It has no table of already-declared conformant members; every file it walks is judged the same way, old or new. So this is not a third instance of #16828; it is a first instance of a blind spot in the auditor, which is a different and arguably cheaper thing to fix.

Two candidate fixes — ⛔ not a ruling, and both want the post-#17048 tree

1. Widen the recognition. Teach POPULATION_CONSTANT (or the surrounding extraction) the object-property shape, so roots: [...] inside a config literal is seen. Fixes the class, not the instance; ⚠️ it will also surface however many other gates spell their population that way, which is a population nobody has counted — see below.
2. Rename the declaration in check-doc-route-spelling.mjs to a recognised shape (const SCAN_ROOTS = […]). Fixes the one instance, cheap, and leaves the auditor as blind as it was.

⭐ (1) is the one that removes the blind spot; (2) is the one that makes today green. ⚠️ Whoever takes it should say which problem they are solving before choosing.

⚠️ The unmeasured half, which decides how big (1) is

Nobody has counted how many gates declare a population as a lowercase property rather than an uppercase const. #17050's sweep could not have told us — it keyed on tree-walk signatures, not declaration shapes. ⛔ A zero from any sweep here needs a positive control that re-finds check-doc-route-spelling.mjs's own roots:, or it is NOT MEASURED, not a clean tree.

Disposition — filed now, against the delivering seat's recommendation, and why

The delivering seat recommended leaving this for whoever next touches bare-root-worklist.mjs after #17048 lands, reasoning that filing now risks a card whose fix design goes stale.

⛔ Overridden, and the reason is a standing correction in this lane's own ledger: a "waiting for someone else" list rots silently. The seat's staleness concern is real but applies to the fix design, not to the fact — and the fact is measured, verified twice, and does not move when #17048 lands. So the fact is recorded here now; the design is explicitly deferred.

Unblock condition, written to be executed rather than interpreted: #17048 merges or closes. It changes the exact predicates bare-root-worklist.mjs imports from dispatch-gates.mjs, so any fix designed before then is designed against a tree about to move. At that moment this card returns to pm:queue and both candidate fixes above are re-derivable against the merged tree.

Provenance

#17050's per-script triage (report issuecomment-5596...), row 8 of 12. That triage's headline result was 0 of 12 rows matching #16828's class — this row is the only genuinely new finding in the twelve, and it is a different species from the one being hunted.

Related: #16828 (the class the sweep was hunting) · #17050 (the sweep) · #17032 · #16770 — all about the distance between what a check is named or shaped like and what it actually reads.

[os-trump]

Claim: session session_012zTkyNHJ7TkuN2oXtP5x37 · branch claude/issue-17057-doc-route-spelling-population-shape · seat domain:services PM (#6021)

Clause-②: no — the whole diff lives in repo-local scripts/**, which no package publishes (root package.json is private: true; no package's files array names scripts/, verified on the tree). ⚠️ STOP-and-report if the delivery finds itself adding an exported symbol to a published package, a new key on a published payload, or any packages/spec/src/** path — that flips Clause-② and this card is not carrying the review tier.

Unblock condition met, verified on the tree — not from a report

#17048 merged as a886fe2d9 (git log origin/main --grep='(#17048)'). pm:blocked → pm:dispatched. The fix design may now be derived against the merged tree, as the card's own unblock sentence specifies.

Premise re-verified by this seat against today's origin/main (91f65c4ea), post-#17048

Both halves of the card still read exactly as filed:

• scripts/check-doc-route-spelling.mjs:131 → roots: ['content/docs', 'skills'], inside const REAL_CONFIG = {
• scripts/pm/bare-root-worklist.mjs:119 → const POPULATION_CONSTANT = /^(?:[A-Z0-9_]*_ROOTS?|[A-Z0-9_]*_DIRS?|ROOTS|DIRS|POPULATION|[A-Z0-9_]*_SCOPE)$/;

⇒ the lowercase property cannot match the uppercase-only name regex. The card's fact did not move when #17048 landed, as it predicted.

The ruling the card deferred: take fix (1), widen the recognition

The card refused to rule and asked whoever takes it to say which problem they are solving first. Ruling now, so it is not re-litigated:

⭐ Fix (2) — renaming roots: to const SCAN_ROOTS — is refused as the whole answer. It makes today green while leaving the auditor's output silently under-counted by an unknown amount. That is precisely "a zero from a probe that cannot answer yes is NOT MEASURED", one level up: bare-root-worklist.mjs would keep reporting a number that is wrong in a direction nobody can see. This lane spent today deleting exactly that shape (#16919, #17039 → #17055, #16828 → #17048); ⛔ do not re-introduce it here.

Fix (1) is the deliverable. Fix (2) is acceptable in addition, if renaming also makes the instance clearer — but never instead.

⚠️ Measure the tail BEFORE committing to scope — this is the part that can go wrong

Widening recognition will surface every other gate that spells its population as a lowercase property, and each surfaced row then needs a disposition (declared / REFUSE-WIDE / fixed) or the gate goes red. Nobody has counted that population. So:

1. Measure first. Widen the recognition, then run the auditor and report the count of newly-surfaced rows before deciding what the PR carries.
2. ⛔ A zero, or a small number, is not a result without a positive control. The control is fixed and named by the card: the sweep must re-find check-doc-route-spelling.mjs's own roots: under the widened recognition. If it does not, the instrument is broken and every number it produced is NOT MEASURED — report that, do not proceed.
3. A crude first cut from this seat, offered as a lead and ⛔ explicitly NOT a census: a line-anchored grep for ^\s*(roots|dirs|scanRoots|searchRoots|population|scope)\s*:\s*\[ over scripts/** returns 22 hits, but nearly all are inside self-test fixture literals (check-type-check-coverage.mjs, check-self-test-wired.mjs, check-self-test-workflow-commands.mjs); the only production-config instance it finds is check-doc-route-spelling.mjs itself (lines 131 and 787). ⚠️ That grep is anchored and name-listed, so it under-counts by construction — treat it as a hint that the tail may be small, never as the measurement. Your instrument must be the widened auditor, not my grep.

Scope gate, written to be executed: if the honest count of newly-surfaced production rows is ≤ 3, carry the widening and their dispositions in this one PR. If it is more, ⛔ stop and report the count with the row list — I will re-scope rather than have you disposition a population you had to guess at.

Fences

• ⛔ Governed surface: skills/** is governed. This gate scans skills/**; the fix must not edit anything under it. If a surfaced row's only fix is inside skills/**, docs/adr/**, .claude/**, AGENTS.md or CLAUDE.md, that is the maintainer's floor — report it, do not write it.
• ⛔ Never skip, disable or quarantine a test to get green.
• ⛔ Never git stash — the stash stack is shared across all worktrees. Use a patch or a wip commit.
• Worktree-first: objectstack-issue-17057, dedicated. ⛔ Never edit the shared primary checkout.
• Single-writer: only .objectui-sha is gate-protected (verified on the tree — scripts/check-single-claim-paths.mjs's SINGLE_CLAIM_PATHS has exactly one entry). ⚠️ scripts/pm/bare-root-worklist.mjs is not a claimed path and no fence applies to it; check for an in-flight PR touching it as ordinary courtesy, not as a gate.
• Commit, push and open the draft PR as soon as there is anything worth showing — before verification finishes. Verification results get appended to the body as they arrive. ⛔ Do not end a turn waiting on a background job or a watcher; nothing wakes a seat that is waiting.

---

Generated by Claude Code