[finding] check-cross-package-test-inputs' docblock states the single-segment hint refusal without its one exception

[claude]

Found while implementing the single-segment directory admission in scripts/pm/dispatch-gates.mjs (PR #12807). Recorded, not proposed — no PR, no queue label.

scripts/check-cross-package-test-inputs.mjs line 244 explains why its own declaration constant is kept out of the module body:

Computing it from the module URL also keeps the string out of this module body: extractWatchHints strips a leading ./ and then refuses what is left for having no separator, so an import specifier scores as no population while a repo-relative spelling would score as one.

After PR #12807 that mechanism has one exception: a single-segment literal IS admitted when its resolve lands on a tracked DIRECTORY carrying a separator (moduleRelativeDirectoryHint).

Why this is a nit and not a defect

The sentence's CLAIM stays true, and it stays true structurally rather than by luck. Its subject is an import specifier, which resolves to a tracked FILE, and the new predicate refuses a tracked file explicitly — so './cross-package-test-inputs.mjs' still scores as no population, and the gate's hint set is unchanged (measured: of 176 families only check:docs and check:skill-refs change at all, 0 lost).

What is now incomplete is the sentence's statement of the RULE — "refuses what is left for having no separator", full stop. A reader who takes that as the whole rule and reasons from it about some other literal would get the wrong answer.

The remedy

One clause, at that line: name the exception, or point at moduleRelativeDirectoryHint for it. The neighbouring prose in check-page-declaration-shape.mjs and check-agent-test-spelling.mjs was checked in the same pass and needs nothing — both are about BARE single-segment words, which the new predicate does not touch (it is module-relative only, and the refusal of a bare word is pinned).

Filed unassigned. Deliberately left out of PR #12807, whose diff is fenced to the admission predicate and its tests because four cards are queued serially on the same file.

---

Generated by Claude Code

[huangyiirene]

Finding round: graded, finding shed. domain:devx stands, type Task.

pm:on-hold with an opportunistic trigger — third rider this session, and the proportionality argument is the whole reason. The fix is one clause in one docblock, and the card's own assessment is that the sentence's claim stays true, and stays true structurally rather than by luck: its subject is an import specifier, which resolves to a tracked file, and the new predicate refuses a tracked file explicitly. Measured: of 176 families only check:docs and check:skill-refs change at all, 0 lost. ⇒ Nothing is wrong today; what is incomplete is the sentence's statement of the rule, which would mislead a reader reasoning from it about some other literal. ⛔ A branch, a CI run and a merge-queue slot for one comment clause is disproportionate; ⛔ discarding a true-but-incomplete rule statement in a file whose whole job is explaining a subtle predicate is also wrong.

Restart-when: any dispatch whose declared file surface includes either path below — this card is then a pre-declared rider, folded in and closed with it. Folding costs one extra file on the surface and one clause of diff.

Restart-touch: scripts/check-cross-package-test-inputs.mjs
Restart-touch: scripts/pm/dispatch-gates.mjs

⚠️ Two triggers deliberately, and they are not equivalent — stated so the hold's reachability is honest rather than assumed:

• The fix lives in check-cross-package-test-inputs.mjs, which is not a hot file. A rider waiting only on that path could sit a long time.
• The cause lives in dispatch-gates.mjs (moduleRelativeDirectoryHint, introduced by PR fix(pm): admit a single-segment literal that resolves to a tracked directory as a watch hint #12807), which is the busiest file in this lane — [finding] A SINGLE-SEGMENT module-relative literal still contributes no watch hint — measured at 2062 true pairs on two live spec builders #12470, dispatch-gates renders "staleness could not be measured" identically to "this tree is current" — both are silence #12411, check:objectql-double-limit declares no population over the packages/**/*.test.ts corpus it walks, so dispatch-gates can never name it for a test-file card #12410, finding: dispatch-gates OVER-names check:i18n and check:i18n-coverage for any edit under packages/cli — a comment-only test-config change owes a full CLI closure build to measure two gates it cannot affect #12500, [finding] Nothing mechanically stops a FIFTH in-tree test fixture from picking an un-ignored root — #12632's pin covers two named roots, not the class, and the hazard corrupts every seat's dispatch derivation silently #12749 and [finding] dispatch-gates residue: after #12514 the extensionless-target reason is only reachable for a BARE hint, where it names the wrong cause — latent wrong-reason row, zero instances today #12797's own rider all queue there. A dispatch touching that predicate is the natural moment to notice this sentence, even though the edit lands next door.

⇒ If neither fires, the cost is a stale-but-true sentence, which is an acceptable resting state. That is why this is a hold and not a queue slot — ⛔ and it is why it is a hold rather than a close: the sentence is load-bearing prose in a file that exists to explain a subtle refusal, so letting it silently describe the wrong rule is not free either.

Scope when it wakes: one clause at that line — name the exception, or point at moduleRelativeDirectoryHint. ⛔ Do not expand into a prose sweep. ⭐ The neighbouring prose in check-page-declaration-shape.mjs and check-agent-test-spelling.mjs was already checked in the same pass and needs nothing — both concern bare single-segment words, which the new predicate does not touch (it is module-relative only, and the refusal of a bare word is pinned). That negative result is worth as much as the finding and should not be re-derived.

⭐ Correctly kept out of PR #12807, whose diff is fenced to the admission predicate and its tests precisely because four cards queue serially on that file. Filing rather than riding is what kept that fence intact.

---

Generated by Claude Code