Filed unassigned while measuring #12173 (finding ①). Not claimed, not graded.
What was measured
scripts/check-objectql-double-limit.mjs walks packages/**/*.test.ts:
const SCAN_ROOT = 'packages'; // :231
...
else if (/\.test\.ts$/.test(entry)) acc.push(p); // :714
for (const abs of testFilesUnder(join(repoRoot, SCAN_ROOT))) { ... } // :727
The only path literals extractWatchHints recovers from it are scripts/objectql-double-limit.baseline.json and n/a — the bare single-segment 'packages' is not recovered at all, and would not cover a nested path even if it were:
$ node -e "... m.extractWatchHints(src)"
hints: ["scripts/objectql-double-limit.baseline.json","n/a"]
hintCovers('packages','packages/objectql/src/a.test.ts') -> false
hintCovers('packages/**','packages/objectql/src/a.test.ts') -> true
Consequence, on a fully current tree (origin/main = b000ab59bb), for a test-file card:
$ node scripts/pm/dispatch-gates.mjs 'packages/plugins/plugin-email/test/send.test.ts'
# 175 families discovered; check:objectql-double-limit absent from "Local gates for this card"
$ node scripts/pm/dispatch-gates.mjs --residue 'packages/plugins/plugin-email/test/send.test.ts'
- pnpm check:objectql-double-limit [lint.yml] names: scripts/objectql-double-limit.baseline.json, n/a
It is in the residue, not the family. Nor is it in CHANGE_KIND_GATES — its two nearest siblings, check:engine-double-contract and check:where-matcher, which walk the same *.test.ts corpus for the same class of fake-engine double, both are under the adds or edits a test file kind.
Why this matters
#12173 records a dev being broadsided on check:objectql-double-limit and attributes the miss to branch-tree staleness (the gate landed on main after the branch was cut). That attribution is incomplete: on today's tree, with the gate fully present, the derivation still cannot name it for a test-file change. Deriving the family list against the merge base would move it from "not discovered" to "discovered, then dropped into residue" — the dispatch prompt would still not carry it.
The fix this repo has already decided
The tool's own residue note prescribes it, verbatim:
Read the gate before treating it as one. If it scans packages, the fix belongs there: declare the scan surface beside the roster (the subtree spelling, packages/**), after which it is MATCHED here.
Same class as #10790 (check:parse-guard, bare scripts refused as too generic) and #10648 (check:doc-authoring, four bare-word population roots), both closed by declaring the subtree spelling in the gate.
⚠️ Whoever takes this should re-derive the residue after the change and check the opposite direction too: packages/** matches ~every package card, so the cost of declaring it is that this gate joins a very large number of families. If that cost is judged too high, the CHANGE_KIND_GATES route (beside its two siblings under adds or edits a test file) is the alternative — the two routes produce different maintenance shapes and the choice is a judgment call, not a mechanical one.
Also observed while here
Lint & Repo Gates still runs 120 named steps sequentially with no continue-on-error key anywhere in the job. Re-derived at b000ab59bb: check:where-matcher at lint.yml:2768, the ObjectQL double limit gate step immediately after it at :2789, and 10 further gate steps in that job after it — #12173 recorded 46 at the time of filing. That half is #12173's finding ②; recorded here only because the number moved.
Filed unassigned while measuring #12173 (finding ①). Not claimed, not graded.
What was measured
scripts/check-objectql-double-limit.mjswalkspackages/**/*.test.ts:The only path literals
extractWatchHintsrecovers from it arescripts/objectql-double-limit.baseline.jsonandn/a— the bare single-segment'packages'is not recovered at all, and would not cover a nested path even if it were:Consequence, on a fully current tree (
origin/main=b000ab59bb), for a test-file card:It is in the residue, not the family. Nor is it in
CHANGE_KIND_GATES— its two nearest siblings,check:engine-double-contractandcheck:where-matcher, which walk the same*.test.tscorpus for the same class of fake-engine double, both are under theadds or edits a test filekind.Why this matters
#12173 records a dev being broadsided on
check:objectql-double-limitand attributes the miss to branch-tree staleness (the gate landed onmainafter the branch was cut). That attribution is incomplete: on today's tree, with the gate fully present, the derivation still cannot name it for a test-file change. Deriving the family list against the merge base would move it from "not discovered" to "discovered, then dropped into residue" — the dispatch prompt would still not carry it.The fix this repo has already decided
The tool's own residue note prescribes it, verbatim:
Same class as #10790 (
check:parse-guard, barescriptsrefused as too generic) and #10648 (check:doc-authoring, four bare-word population roots), both closed by declaring the subtree spelling in the gate.packages/**matches ~every package card, so the cost of declaring it is that this gate joins a very large number of families. If that cost is judged too high, theCHANGE_KIND_GATESroute (beside its two siblings underadds or edits a test file) is the alternative — the two routes produce different maintenance shapes and the choice is a judgment call, not a mechanical one.Also observed while here
Lint & Repo Gatesstill runs 120 named steps sequentially with nocontinue-on-errorkey anywhere in the job. Re-derived atb000ab59bb:check:where-matcheratlint.yml:2768, theObjectQL double limit gatestep immediately after it at:2789, and 10 further gate steps in that job after it — #12173 recorded 46 at the time of filing. That half is #12173's finding ②; recorded here only because the number moved.