Found while flooring scripts/workspace-enumerator.mjs for #15375. Filed unassigned, out of that card's scope.
The reading
scripts/measure-self-test-floor.mjs selects its population on a --self-test dispatch, and spells that criterion as an argv-membership test:
/** A `--self-test` DISPATCH: argv membership tested, not a literal passed to a child. */
const DISPATCH = /(?:includes|has)\(\s*['"`]--self-test['"`]\s*\)/;
Only includes(...) and has(...) match. A script that dispatches by equality on an already-extracted flag does not:
// scripts/pnpm-filter-targets.mjs:695
if (flag === '--self-test') process.exit(await selfTest());
That is a real dispatch with a real CI invocation — check:pnpm-filter-targets runs node scripts/pnpm-filter-targets.mjs --self-test in lint.yml — and the census still cannot see the file. Measured on b3cc2cd57: the instrument reports 170 files and pnpm-filter-targets.mjs is not among them, under any classification.
Scope, measured
Six tracked scripts/** files dispatch on --self-test by equality and match neither arm of DISPATCH:
scripts/check-closing-keyword-parity.mjsscripts/check-pnpm-filter-targets.mjsscripts/check-settings-bind-window.mjsscripts/check-step-collectors.mjsscripts/measure-durability-swallow-family.mjsscripts/pnpm-filter-targets.mjs
Their absence from the census is not a clearance — as with #15375, it is a set of files the criterion cannot reach, so no batch of #13799 will be dispatched against them and nothing records that they were never floored.
Boundary — what this is NOT
This is a different hole from #15375, which reports a module with no dispatch at all and whose header argues it must never acquire one. These six DO dispatch, and five of the six are gates CI already invokes with --self-test; only the criterion's spelling keeps them out. It is also distinct from #15153 (dispatch routed through a MODES table) and from #15339 (PRODUCES_FAILURE and a ternary process.exit), both of which are about other predicates in the same instrument.
Whether the repair is to widen DISPATCH, to add these as ENTRY_BY_HAND rows, or to leave the criterion alone and record the six as a known blind spot is the instrument's own question — the same standing question as #15391 / #15371 / #14963. Not chosen here.
Reproduce
node scripts/measure-self-test-floor.mjs --json | python3 -c "import json,sys; d=json.load(sys.stdin); print(len(d)); print([r for r in d if 'pnpm-filter-targets' in r['file']])"
grep -n "flag === '--self-test'" scripts/pnpm-filter-targets.mjs
Generated by Claude Code
Found while flooring
scripts/workspace-enumerator.mjsfor #15375. Filed unassigned, out of that card's scope.The reading
scripts/measure-self-test-floor.mjsselects its population on a--self-testdispatch, and spells that criterion as an argv-membership test:Only
includes(...)andhas(...)match. A script that dispatches by equality on an already-extracted flag does not:That is a real dispatch with a real CI invocation —
check:pnpm-filter-targetsrunsnode scripts/pnpm-filter-targets.mjs --self-testin lint.yml — and the census still cannot see the file. Measured onb3cc2cd57: the instrument reports 170 files andpnpm-filter-targets.mjsis not among them, under any classification.Scope, measured
Six tracked
scripts/**files dispatch on--self-testby equality and match neither arm ofDISPATCH:scripts/check-closing-keyword-parity.mjsscripts/check-pnpm-filter-targets.mjsscripts/check-settings-bind-window.mjsscripts/check-step-collectors.mjsscripts/measure-durability-swallow-family.mjsscripts/pnpm-filter-targets.mjsTheir absence from the census is not a clearance — as with #15375, it is a set of files the criterion cannot reach, so no batch of #13799 will be dispatched against them and nothing records that they were never floored.
Boundary — what this is NOT
This is a different hole from #15375, which reports a module with no dispatch at all and whose header argues it must never acquire one. These six DO dispatch, and five of the six are gates CI already invokes with
--self-test; only the criterion's spelling keeps them out. It is also distinct from #15153 (dispatch routed through a MODES table) and from #15339 (PRODUCES_FAILUREand a ternaryprocess.exit), both of which are about other predicates in the same instrument.Whether the repair is to widen
DISPATCH, to add these asENTRY_BY_HANDrows, or to leave the criterion alone and record the six as a known blind spot is the instrument's own question — the same standing question as #15391 / #15371 / #14963. Not chosen here.Reproduce
Generated by Claude Code