fix(devx): dispatch-gates marks dead watch-hint literals in reachable families and refuses @-scoped specifiers as paths (#13312) by os-project-manager · Pull Request #13447 · objectstack-ai/objectstack

os-project-manager · 2026-08-30T12:27:18Z

The defect, and the measurement that chose the route

dispatch-gates --residue printed three literals in check:query-options-erasure's names: line that name files which have never existed in this tree, verbatim, with no dead annotation — the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled.

The card declared its scope unmeasured and listed three routes (extractor / annotator / gate-rename) without deciding. Per the dispatch ruling I re-derived the population across ALL families first, on this branch's base 5f0a9c4ad:

reading	before	after (this PR, head `a3d95518`)
families discovered	187	187
families declaring a population	151	149 (2 moved to honest `undetermined`)
reachable families carrying dead literals	65	49
dead literals riding unannotated in reachable families	598	0 (245 remain, every one marked and counted)
fully-dead families (annotated before and after)	3	1

Shape split of the 598: 353 were @-scoped package specifiers (@objectstack/spec, @fx/canary, @typescript-eslint/parser, @/lib/i18n) — the widest single species, exactly the one the third-instance comment on the card measured. 245 are path-shaped fixture names, owner/repo slugs, action specs and drifted doc examples — no closed shape can refuse those, so they are annotated instead of removed.

The measurement picked the route: gate-rename was dead on arrival (it fixes 3 of 598), and the numbers chose one narrow extractor refusal plus the unconditional annotator.

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Annotator (unconditional per triage).deadHintSweep is the existing reachability sweep at per-hint grain; unreachableFamilies now reads its answer off it (one sweep, two grains, no disagreement — pinned). Under --residue, a dead literal is marked where shown and counted where not:

- pnpm check:query-options-erasure [lint.yml] names: scripts/query-options-erasure-baseline.json, scripts/__closure_fixture_gate__.mjs X, scripts/__closure_fixture_helper__.mjs X, ...
3 of 8 declared literal(s) reach nothing tracked (marked X where shown) - dead leads, not population: ...

(The live output uses the ballot-X mark; spelled as X here to keep this body free of markup-sensitive characters.) Display only — no verdict moves; matched/undetermined/silent and the unreachable listing are byte-compatible except for the new marks and notes.

Extractor: isNonPathNamespace gains a third closed shape. A first segment beginning with @ is a scope marker, not a directory — npm grammar, not a heuristic. Zero tracked paths have an @-headed first segment (measured at 5f0a9c4ad), so all 353 refused literals were dead: no live hint moves, no card's verdict moves. The two families whose ENTIRE population was a package-name ledger (check:driver-memory-census, check-test-completeness.mjs — the third-instance comment's table) leave the unreachable listing for undetermined, the bucket that predicate's docblock calls strictly more honest. check:stall-guard-headroom's owner/repo slug is NOT refusable by shape (two bare words are what a path looks like — pinned in that direction) and is now annotated instead.

Zone 2 hypothesis: falsified, with numbers

The PM's hypothesis was that the extractor should stop scraping self-test fixture constants because "a fixture-name constant is not a declaration in any sense." I prototyped exactly that (mask module-scope constants whose identifier is referenced only inside self-test bodies, iterated to a fixed point) and measured it: 96 hints removed — 10 dead, 86 LIVE. The live removals include packages/** and skills/** subtree declarations, whole published-files rosters, and baseline artifacts — because this repo's sanctioned declaration idiom (ROOT_DIR_WATCH_HINTS: a module-scope constant that exists FOR the scanner, referenced only by the self-test that pins it) is structurally indistinguishable from a fixture constant. Reference analysis deletes 8.6 real populations per fabrication removed. In this codebase a scanner-facing constant IS a declaration, so the premise fails and that route is refused. The fixture-name class stays in the hint sets and is now annotated.

The three non-empty delivery controls (from triage)

A real hint still prints unmarked: scripts/query-options-erasure-baseline.json leads the family's names: line above; pinned as ...and leaves the real hint beside it unmarked.
A planted dead literal in a reachable family IS annotated: pinned — scripts/__planted_never_existed__.mjs planted beside a live hint is swept, marked in the names fragment, and counted 1 of 2 with its reason in the note.
Population numbers: table above; 65 of 151 declaring families carried 598 unannotated dead literals, after the fix 0 unannotated (353 refused at extraction, 245 annotated).

Verification (all at head `a3d95518`, exit codes captured before any pipe)

pnpm check:pm-dispatch-gates — verdict line: dispatch-gates self-test: 944 cases pass. (includes the ~30 new pins; one pre-existing pin's specimen-finder needed a third exclusion — the job-filter key, #12956's channel — because thinning the inherited pairs made its .find land on a job-filtered family first; the pin's claim is unchanged)
node scripts/pm/bare-root-worklist.mjs --self-test — OK self-test: 51 live row(s), 43 unreachable as spelled, 43 recorded verdict(s) - none stale, none missing, none contradicted.
node scripts/check-self-test-wired.mjs — every one of the 151 script(s) CI runs that ship a --self-test has that self-test run by CI.
Derived gate union re-run on the final head (dispatch-gates no-path mode, banner names commit a3d95518): check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:entry-guard, check:parse-guard, check:pm-dispatch-gates, check:pnpm-filter-targets, check:watch-hint-literal, check-ci-filter-parity, check-self-test-wired, plus check:nul-bytes — all exit 0, each verdict read from the gate's own judgment line.
NOT run locally (declared narrowing): check-shard-attestation.mjs, check-test-completeness.mjs, check-cross-package-test-inputs.mjs in their ci.yml forms — they read CI run artifacts (test logs) that do not exist locally; CI runs them on this PR.

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

hintCovers does not follow a final-segment basename glob: .changeset/*.md covers zero tracked files while the tree holds ~400 matching changesets, so several changeset gates' hints measure dead and the reason renderer calls it layout moved, which is false. Verdict-moving to fix; not this card.
check:doc-anchors carries the fabricated hint scripts/scripts/check-doc-anchors.mjs: line 459 of that gate spells ./scripts/check-doc-anchors.mjs inside a remedy-message string meant to run from the repo root, and the module-relative resolve doubles the prefix. Now annotated; the spelling fix belongs to that gate.

This PR is skip-changeset: the whole diff is scripts/pm/dispatch-gates.mjs, internal PM tooling — nothing is published from a package.

Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

🤖 Generated with Claude Code

Generated by Claude Code

…fuse @-scoped specifiers as paths Three of the five hints --residue showed for check:query-options-erasure named files that have never existed, printed verbatim: the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled. Measured across all families before choosing the route: 65 of 151 declaring families carried 598 such literals, 353 of them @-scoped package specifiers. - deadHintSweep: the reachability sweep at per-hint grain; the unreachable listing and the new residue annotations read one sweep, two grains. - residueNames / deadNamesNote: dead literals are marked where shown and counted with reasons where not. - isNonPathNamespace: an @-headed first segment (npm scope, version-suffixed specifier, bundler alias) is a namespace, not a path — 353 dead hints refused at extraction, zero tracked paths start with @, no live hint moves. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

os-project-manager · 2026-08-30T12:27:18Z

Fixes#13312

The defect, and the measurement that chose the route

dispatch-gates --residue printed three literals in check:query-options-erasure's names: line that name files which have never existed in this tree, verbatim, with no dead annotation — the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled.

The card declared its scope unmeasured and listed three routes (extractor / annotator / gate-rename) without deciding. Per the dispatch ruling I re-derived the population across ALL families first, on this branch's base 5f0a9c4ad:

reading	before	after (this PR, head `a3d95518`)
families discovered	187	187
families declaring a population	151	149 (2 moved to honest `undetermined`)
reachable families carrying dead literals	65	49
dead literals riding unannotated in reachable families	598	0 (245 remain, every one marked and counted)
fully-dead families (annotated before and after)	3	1

Shape split of the 598: 353 were @-scoped package specifiers (@objectstack/spec, @fx/canary, @typescript-eslint/parser, @/lib/i18n) — the widest single species, exactly the one the third-instance comment on the card measured. 245 are path-shaped fixture names, owner/repo slugs, action specs and drifted doc examples — no closed shape can refuse those, so they are annotated instead of removed.

The measurement picked the route: gate-rename was dead on arrival (it fixes 3 of 598), and the numbers chose one narrow extractor refusal plus the unconditional annotator.

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Annotator (unconditional per triage).deadHintSweep is the existing reachability sweep at per-hint grain; unreachableFamilies now reads its answer off it (one sweep, two grains, no disagreement — pinned). Under --residue, a dead literal is marked where shown and counted where not:

- pnpm check:query-options-erasure [lint.yml] names: scripts/query-options-erasure-baseline.json, scripts/__closure_fixture_gate__.mjs X, scripts/__closure_fixture_helper__.mjs X, ...
3 of 8 declared literal(s) reach nothing tracked (marked X where shown) - dead leads, not population: ...

(The live output uses the ballot-X mark; spelled as X here to keep this body free of markup-sensitive characters.) Display only — no verdict moves; matched/undetermined/silent and the unreachable listing are byte-compatible except for the new marks and notes.

Extractor: isNonPathNamespace gains a third closed shape. A first segment beginning with @ is a scope marker, not a directory — npm grammar, not a heuristic. Zero tracked paths have an @-headed first segment (measured at 5f0a9c4ad), so all 353 refused literals were dead: no live hint moves, no card's verdict moves. The two families whose ENTIRE population was a package-name ledger (check:driver-memory-census, check-test-completeness.mjs — the third-instance comment's table) leave the unreachable listing for undetermined, the bucket that predicate's docblock calls strictly more honest. check:stall-guard-headroom's owner/repo slug is NOT refusable by shape (two bare words are what a path looks like — pinned in that direction) and is now annotated instead.

Zone 2 hypothesis: falsified, with numbers

The PM's hypothesis was that the extractor should stop scraping self-test fixture constants because "a fixture-name constant is not a declaration in any sense." I prototyped exactly that (mask module-scope constants whose identifier is referenced only inside self-test bodies, iterated to a fixed point) and measured it: 96 hints removed — 10 dead, 86 LIVE. The live removals include packages/** and skills/** subtree declarations, whole published-files rosters, and baseline artifacts — because this repo's sanctioned declaration idiom (ROOT_DIR_WATCH_HINTS: a module-scope constant that exists FOR the scanner, referenced only by the self-test that pins it) is structurally indistinguishable from a fixture constant. Reference analysis deletes 8.6 real populations per fabrication removed. In this codebase a scanner-facing constant IS a declaration, so the premise fails and that route is refused. The fixture-name class stays in the hint sets and is now annotated.

The three non-empty delivery controls (from triage)

A real hint still prints unmarked: scripts/query-options-erasure-baseline.json leads the family's names: line above; pinned as ...and leaves the real hint beside it unmarked.
A planted dead literal in a reachable family IS annotated: pinned — scripts/__planted_never_existed__.mjs planted beside a live hint is swept, marked in the names fragment, and counted 1 of 2 with its reason in the note.
Population numbers: table above; 65 of 151 declaring families carried 598 unannotated dead literals, after the fix 0 unannotated (353 refused at extraction, 245 annotated).

Verification (all at head `a3d95518`, exit codes captured before any pipe)

pnpm check:pm-dispatch-gates — verdict line: dispatch-gates self-test: 944 cases pass. (includes the ~30 new pins; one pre-existing pin's specimen-finder needed a third exclusion — the job-filter key, #12956's channel — because thinning the inherited pairs made its .find land on a job-filtered family first; the pin's claim is unchanged)
node scripts/pm/bare-root-worklist.mjs --self-test — OK self-test: 51 live row(s), 43 unreachable as spelled, 43 recorded verdict(s) - none stale, none missing, none contradicted.
node scripts/check-self-test-wired.mjs — every one of the 151 script(s) CI runs that ship a --self-test has that self-test run by CI.
Derived gate union re-run on the final head (dispatch-gates no-path mode, banner names commit a3d95518): check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:entry-guard, check:parse-guard, check:pm-dispatch-gates, check:pnpm-filter-targets, check:watch-hint-literal, check-ci-filter-parity, check-self-test-wired, plus check:nul-bytes — all exit 0, each verdict read from the gate's own judgment line.
NOT run locally (declared narrowing): check-shard-attestation.mjs, check-test-completeness.mjs, check-cross-package-test-inputs.mjs in their ci.yml forms — they read CI run artifacts (test logs) that do not exist locally; CI runs them on this PR.

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

hintCovers does not follow a final-segment basename glob: .changeset/*.md covers zero tracked files while the tree holds ~400 matching changesets, so several changeset gates' hints measure dead and the reason renderer calls it layout moved, which is false. Verdict-moving to fix; not this card.
check:doc-anchors carries the fabricated hint scripts/scripts/check-doc-anchors.mjs: line 459 of that gate spells ./scripts/check-doc-anchors.mjs inside a remedy-message string meant to run from the repo root, and the module-relative resolve doubles the prefix. Now annotated; the spelling fix belongs to that gate.

This PR is skip-changeset: the whole diff is scripts/pm/dispatch-gates.mjs, internal PM tooling — nothing is published from a package.

Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

🤖 Generated with Claude Code

Generated by Claude Code

…fuse @-scoped specifiers as paths Three of the five hints --residue showed for check:query-options-erasure named files that have never existed, printed verbatim: the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled. Measured across all families before choosing the route: 65 of 151 declaring families carried 598 such literals, 353 of them @-scoped package specifiers. - deadHintSweep: the reachability sweep at per-hint grain; the unreachable listing and the new residue annotations read one sweep, two grains. - residueNames / deadNamesNote: dead literals are marked where shown and counted with reasons where not. - isNonPathNamespace: an @-headed first segment (npm scope, version-suffixed specifier, bundler alias) is a namespace, not a path — 353 dead hints refused at extraction, zero tracked paths start with @, no live hint moves. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

os-project-manager · 2026-08-30T12:27:18Z

Fixes#13312

The defect, and the measurement that chose the route

dispatch-gates --residue printed three literals in check:query-options-erasure's names: line that name files which have never existed in this tree, verbatim, with no dead annotation — the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled.

The card declared its scope unmeasured and listed three routes (extractor / annotator / gate-rename) without deciding. Per the dispatch ruling I re-derived the population across ALL families first, on this branch's base 5f0a9c4ad:

reading	before	after (this PR, head `a3d95518`)
families discovered	187	187
families declaring a population	151	149 (2 moved to honest `undetermined`)
reachable families carrying dead literals	65	49
dead literals riding unannotated in reachable families	598	0 (245 remain, every one marked and counted)
fully-dead families (annotated before and after)	3	1

Shape split of the 598: 353 were @-scoped package specifiers (@objectstack/spec, @fx/canary, @typescript-eslint/parser, @/lib/i18n) — the widest single species, exactly the one the third-instance comment on the card measured. 245 are path-shaped fixture names, owner/repo slugs, action specs and drifted doc examples — no closed shape can refuse those, so they are annotated instead of removed.

The measurement picked the route: gate-rename was dead on arrival (it fixes 3 of 598), and the numbers chose one narrow extractor refusal plus the unconditional annotator.

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Annotator (unconditional per triage).deadHintSweep is the existing reachability sweep at per-hint grain; unreachableFamilies now reads its answer off it (one sweep, two grains, no disagreement — pinned). Under --residue, a dead literal is marked where shown and counted where not:

- pnpm check:query-options-erasure [lint.yml] names: scripts/query-options-erasure-baseline.json, scripts/__closure_fixture_gate__.mjs X, scripts/__closure_fixture_helper__.mjs X, ...
3 of 8 declared literal(s) reach nothing tracked (marked X where shown) - dead leads, not population: ...

(The live output uses the ballot-X mark; spelled as X here to keep this body free of markup-sensitive characters.) Display only — no verdict moves; matched/undetermined/silent and the unreachable listing are byte-compatible except for the new marks and notes.

Extractor: isNonPathNamespace gains a third closed shape. A first segment beginning with @ is a scope marker, not a directory — npm grammar, not a heuristic. Zero tracked paths have an @-headed first segment (measured at 5f0a9c4ad), so all 353 refused literals were dead: no live hint moves, no card's verdict moves. The two families whose ENTIRE population was a package-name ledger (check:driver-memory-census, check-test-completeness.mjs — the third-instance comment's table) leave the unreachable listing for undetermined, the bucket that predicate's docblock calls strictly more honest. check:stall-guard-headroom's owner/repo slug is NOT refusable by shape (two bare words are what a path looks like — pinned in that direction) and is now annotated instead.

Zone 2 hypothesis: falsified, with numbers

The PM's hypothesis was that the extractor should stop scraping self-test fixture constants because "a fixture-name constant is not a declaration in any sense." I prototyped exactly that (mask module-scope constants whose identifier is referenced only inside self-test bodies, iterated to a fixed point) and measured it: 96 hints removed — 10 dead, 86 LIVE. The live removals include packages/** and skills/** subtree declarations, whole published-files rosters, and baseline artifacts — because this repo's sanctioned declaration idiom (ROOT_DIR_WATCH_HINTS: a module-scope constant that exists FOR the scanner, referenced only by the self-test that pins it) is structurally indistinguishable from a fixture constant. Reference analysis deletes 8.6 real populations per fabrication removed. In this codebase a scanner-facing constant IS a declaration, so the premise fails and that route is refused. The fixture-name class stays in the hint sets and is now annotated.

The three non-empty delivery controls (from triage)

A real hint still prints unmarked: scripts/query-options-erasure-baseline.json leads the family's names: line above; pinned as ...and leaves the real hint beside it unmarked.
A planted dead literal in a reachable family IS annotated: pinned — scripts/__planted_never_existed__.mjs planted beside a live hint is swept, marked in the names fragment, and counted 1 of 2 with its reason in the note.
Population numbers: table above; 65 of 151 declaring families carried 598 unannotated dead literals, after the fix 0 unannotated (353 refused at extraction, 245 annotated).

Verification (all at head `a3d95518`, exit codes captured before any pipe)

pnpm check:pm-dispatch-gates — verdict line: dispatch-gates self-test: 944 cases pass. (includes the ~30 new pins; one pre-existing pin's specimen-finder needed a third exclusion — the job-filter key, #12956's channel — because thinning the inherited pairs made its .find land on a job-filtered family first; the pin's claim is unchanged)
node scripts/pm/bare-root-worklist.mjs --self-test — OK self-test: 51 live row(s), 43 unreachable as spelled, 43 recorded verdict(s) - none stale, none missing, none contradicted.
node scripts/check-self-test-wired.mjs — every one of the 151 script(s) CI runs that ship a --self-test has that self-test run by CI.
Derived gate union re-run on the final head (dispatch-gates no-path mode, banner names commit a3d95518): check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:entry-guard, check:parse-guard, check:pm-dispatch-gates, check:pnpm-filter-targets, check:watch-hint-literal, check-ci-filter-parity, check-self-test-wired, plus check:nul-bytes — all exit 0, each verdict read from the gate's own judgment line.
NOT run locally (declared narrowing): check-shard-attestation.mjs, check-test-completeness.mjs, check-cross-package-test-inputs.mjs in their ci.yml forms — they read CI run artifacts (test logs) that do not exist locally; CI runs them on this PR.

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

hintCovers does not follow a final-segment basename glob: .changeset/*.md covers zero tracked files while the tree holds ~400 matching changesets, so several changeset gates' hints measure dead and the reason renderer calls it layout moved, which is false. Verdict-moving to fix; not this card.
check:doc-anchors carries the fabricated hint scripts/scripts/check-doc-anchors.mjs: line 459 of that gate spells ./scripts/check-doc-anchors.mjs inside a remedy-message string meant to run from the repo root, and the module-relative resolve doubles the prefix. Now annotated; the spelling fix belongs to that gate.

This PR is skip-changeset: the whole diff is scripts/pm/dispatch-gates.mjs, internal PM tooling — nothing is published from a package.

Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

🤖 Generated with Claude Code

Generated by Claude Code

…fuse @-scoped specifiers as paths Three of the five hints --residue showed for check:query-options-erasure named files that have never existed, printed verbatim: the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled. Measured across all families before choosing the route: 65 of 151 declaring families carried 598 such literals, 353 of them @-scoped package specifiers. - deadHintSweep: the reachability sweep at per-hint grain; the unreachable listing and the new residue annotations read one sweep, two grains. - residueNames / deadNamesNote: dead literals are marked where shown and counted with reasons where not. - isNonPathNamespace: an @-headed first segment (npm scope, version-suffixed specifier, bundler alias) is a namespace, not a path — 353 dead hints refused at extraction, zero tracked paths start with @, no live hint moves. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

os-project-manager · 2026-08-30T12:27:18Z

Fixes#13312

The defect, and the measurement that chose the route

dispatch-gates --residue printed three literals in check:query-options-erasure's names: line that name files which have never existed in this tree, verbatim, with no dead annotation — the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled.

The card declared its scope unmeasured and listed three routes (extractor / annotator / gate-rename) without deciding. Per the dispatch ruling I re-derived the population across ALL families first, on this branch's base 5f0a9c4ad:

reading	before	after (this PR, head `a3d95518`)
families discovered	187	187
families declaring a population	151	149 (2 moved to honest `undetermined`)
reachable families carrying dead literals	65	49
dead literals riding unannotated in reachable families	598	0 (245 remain, every one marked and counted)
fully-dead families (annotated before and after)	3	1

Shape split of the 598: 353 were @-scoped package specifiers (@objectstack/spec, @fx/canary, @typescript-eslint/parser, @/lib/i18n) — the widest single species, exactly the one the third-instance comment on the card measured. 245 are path-shaped fixture names, owner/repo slugs, action specs and drifted doc examples — no closed shape can refuse those, so they are annotated instead of removed.

The measurement picked the route: gate-rename was dead on arrival (it fixes 3 of 598), and the numbers chose one narrow extractor refusal plus the unconditional annotator.

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Annotator (unconditional per triage).deadHintSweep is the existing reachability sweep at per-hint grain; unreachableFamilies now reads its answer off it (one sweep, two grains, no disagreement — pinned). Under --residue, a dead literal is marked where shown and counted where not:

- pnpm check:query-options-erasure [lint.yml] names: scripts/query-options-erasure-baseline.json, scripts/__closure_fixture_gate__.mjs X, scripts/__closure_fixture_helper__.mjs X, ...
3 of 8 declared literal(s) reach nothing tracked (marked X where shown) - dead leads, not population: ...

(The live output uses the ballot-X mark; spelled as X here to keep this body free of markup-sensitive characters.) Display only — no verdict moves; matched/undetermined/silent and the unreachable listing are byte-compatible except for the new marks and notes.

Extractor: isNonPathNamespace gains a third closed shape. A first segment beginning with @ is a scope marker, not a directory — npm grammar, not a heuristic. Zero tracked paths have an @-headed first segment (measured at 5f0a9c4ad), so all 353 refused literals were dead: no live hint moves, no card's verdict moves. The two families whose ENTIRE population was a package-name ledger (check:driver-memory-census, check-test-completeness.mjs — the third-instance comment's table) leave the unreachable listing for undetermined, the bucket that predicate's docblock calls strictly more honest. check:stall-guard-headroom's owner/repo slug is NOT refusable by shape (two bare words are what a path looks like — pinned in that direction) and is now annotated instead.

Zone 2 hypothesis: falsified, with numbers

The PM's hypothesis was that the extractor should stop scraping self-test fixture constants because "a fixture-name constant is not a declaration in any sense." I prototyped exactly that (mask module-scope constants whose identifier is referenced only inside self-test bodies, iterated to a fixed point) and measured it: 96 hints removed — 10 dead, 86 LIVE. The live removals include packages/** and skills/** subtree declarations, whole published-files rosters, and baseline artifacts — because this repo's sanctioned declaration idiom (ROOT_DIR_WATCH_HINTS: a module-scope constant that exists FOR the scanner, referenced only by the self-test that pins it) is structurally indistinguishable from a fixture constant. Reference analysis deletes 8.6 real populations per fabrication removed. In this codebase a scanner-facing constant IS a declaration, so the premise fails and that route is refused. The fixture-name class stays in the hint sets and is now annotated.

The three non-empty delivery controls (from triage)

A real hint still prints unmarked: scripts/query-options-erasure-baseline.json leads the family's names: line above; pinned as ...and leaves the real hint beside it unmarked.
A planted dead literal in a reachable family IS annotated: pinned — scripts/__planted_never_existed__.mjs planted beside a live hint is swept, marked in the names fragment, and counted 1 of 2 with its reason in the note.
Population numbers: table above; 65 of 151 declaring families carried 598 unannotated dead literals, after the fix 0 unannotated (353 refused at extraction, 245 annotated).

Verification (all at head `a3d95518`, exit codes captured before any pipe)

pnpm check:pm-dispatch-gates — verdict line: dispatch-gates self-test: 944 cases pass. (includes the ~30 new pins; one pre-existing pin's specimen-finder needed a third exclusion — the job-filter key, #12956's channel — because thinning the inherited pairs made its .find land on a job-filtered family first; the pin's claim is unchanged)
node scripts/pm/bare-root-worklist.mjs --self-test — OK self-test: 51 live row(s), 43 unreachable as spelled, 43 recorded verdict(s) - none stale, none missing, none contradicted.
node scripts/check-self-test-wired.mjs — every one of the 151 script(s) CI runs that ship a --self-test has that self-test run by CI.
Derived gate union re-run on the final head (dispatch-gates no-path mode, banner names commit a3d95518): check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:entry-guard, check:parse-guard, check:pm-dispatch-gates, check:pnpm-filter-targets, check:watch-hint-literal, check-ci-filter-parity, check-self-test-wired, plus check:nul-bytes — all exit 0, each verdict read from the gate's own judgment line.
NOT run locally (declared narrowing): check-shard-attestation.mjs, check-test-completeness.mjs, check-cross-package-test-inputs.mjs in their ci.yml forms — they read CI run artifacts (test logs) that do not exist locally; CI runs them on this PR.

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

hintCovers does not follow a final-segment basename glob: .changeset/*.md covers zero tracked files while the tree holds ~400 matching changesets, so several changeset gates' hints measure dead and the reason renderer calls it layout moved, which is false. Verdict-moving to fix; not this card.
check:doc-anchors carries the fabricated hint scripts/scripts/check-doc-anchors.mjs: line 459 of that gate spells ./scripts/check-doc-anchors.mjs inside a remedy-message string meant to run from the repo root, and the module-relative resolve doubles the prefix. Now annotated; the spelling fix belongs to that gate.

This PR is skip-changeset: the whole diff is scripts/pm/dispatch-gates.mjs, internal PM tooling — nothing is published from a package.

Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

🤖 Generated with Claude Code

Generated by Claude Code

…fuse @-scoped specifiers as paths Three of the five hints --residue showed for check:query-options-erasure named files that have never existed, printed verbatim: the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled. Measured across all families before choosing the route: 65 of 151 declaring families carried 598 such literals, 353 of them @-scoped package specifiers. - deadHintSweep: the reachability sweep at per-hint grain; the unreachable listing and the new residue annotations read one sweep, two grains. - residueNames / deadNamesNote: dead literals are marked where shown and counted with reasons where not. - isNonPathNamespace: an @-headed first segment (npm scope, version-suffixed specifier, bundler alias) is a namespace, not a path — 353 dead hints refused at extraction, zero tracked paths start with @, no live hint moves. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

os-project-manager · 2026-08-30T12:27:18Z

Fixes#13312

The defect, and the measurement that chose the route

dispatch-gates --residue printed three literals in check:query-options-erasure's names: line that name files which have never existed in this tree, verbatim, with no dead annotation — the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled.

The card declared its scope unmeasured and listed three routes (extractor / annotator / gate-rename) without deciding. Per the dispatch ruling I re-derived the population across ALL families first, on this branch's base 5f0a9c4ad:

reading	before	after (this PR, head `a3d95518`)
families discovered	187	187
families declaring a population	151	149 (2 moved to honest `undetermined`)
reachable families carrying dead literals	65	49
dead literals riding unannotated in reachable families	598	0 (245 remain, every one marked and counted)
fully-dead families (annotated before and after)	3	1

Shape split of the 598: 353 were @-scoped package specifiers (@objectstack/spec, @fx/canary, @typescript-eslint/parser, @/lib/i18n) — the widest single species, exactly the one the third-instance comment on the card measured. 245 are path-shaped fixture names, owner/repo slugs, action specs and drifted doc examples — no closed shape can refuse those, so they are annotated instead of removed.

The measurement picked the route: gate-rename was dead on arrival (it fixes 3 of 598), and the numbers chose one narrow extractor refusal plus the unconditional annotator.

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Annotator (unconditional per triage).deadHintSweep is the existing reachability sweep at per-hint grain; unreachableFamilies now reads its answer off it (one sweep, two grains, no disagreement — pinned). Under --residue, a dead literal is marked where shown and counted where not:

- pnpm check:query-options-erasure [lint.yml] names: scripts/query-options-erasure-baseline.json, scripts/__closure_fixture_gate__.mjs X, scripts/__closure_fixture_helper__.mjs X, ...
3 of 8 declared literal(s) reach nothing tracked (marked X where shown) - dead leads, not population: ...

(The live output uses the ballot-X mark; spelled as X here to keep this body free of markup-sensitive characters.) Display only — no verdict moves; matched/undetermined/silent and the unreachable listing are byte-compatible except for the new marks and notes.

Extractor: isNonPathNamespace gains a third closed shape. A first segment beginning with @ is a scope marker, not a directory — npm grammar, not a heuristic. Zero tracked paths have an @-headed first segment (measured at 5f0a9c4ad), so all 353 refused literals were dead: no live hint moves, no card's verdict moves. The two families whose ENTIRE population was a package-name ledger (check:driver-memory-census, check-test-completeness.mjs — the third-instance comment's table) leave the unreachable listing for undetermined, the bucket that predicate's docblock calls strictly more honest. check:stall-guard-headroom's owner/repo slug is NOT refusable by shape (two bare words are what a path looks like — pinned in that direction) and is now annotated instead.

Zone 2 hypothesis: falsified, with numbers

The PM's hypothesis was that the extractor should stop scraping self-test fixture constants because "a fixture-name constant is not a declaration in any sense." I prototyped exactly that (mask module-scope constants whose identifier is referenced only inside self-test bodies, iterated to a fixed point) and measured it: 96 hints removed — 10 dead, 86 LIVE. The live removals include packages/** and skills/** subtree declarations, whole published-files rosters, and baseline artifacts — because this repo's sanctioned declaration idiom (ROOT_DIR_WATCH_HINTS: a module-scope constant that exists FOR the scanner, referenced only by the self-test that pins it) is structurally indistinguishable from a fixture constant. Reference analysis deletes 8.6 real populations per fabrication removed. In this codebase a scanner-facing constant IS a declaration, so the premise fails and that route is refused. The fixture-name class stays in the hint sets and is now annotated.

The three non-empty delivery controls (from triage)

A real hint still prints unmarked: scripts/query-options-erasure-baseline.json leads the family's names: line above; pinned as ...and leaves the real hint beside it unmarked.
A planted dead literal in a reachable family IS annotated: pinned — scripts/__planted_never_existed__.mjs planted beside a live hint is swept, marked in the names fragment, and counted 1 of 2 with its reason in the note.
Population numbers: table above; 65 of 151 declaring families carried 598 unannotated dead literals, after the fix 0 unannotated (353 refused at extraction, 245 annotated).

Verification (all at head `a3d95518`, exit codes captured before any pipe)

pnpm check:pm-dispatch-gates — verdict line: dispatch-gates self-test: 944 cases pass. (includes the ~30 new pins; one pre-existing pin's specimen-finder needed a third exclusion — the job-filter key, #12956's channel — because thinning the inherited pairs made its .find land on a job-filtered family first; the pin's claim is unchanged)
node scripts/pm/bare-root-worklist.mjs --self-test — OK self-test: 51 live row(s), 43 unreachable as spelled, 43 recorded verdict(s) - none stale, none missing, none contradicted.
node scripts/check-self-test-wired.mjs — every one of the 151 script(s) CI runs that ship a --self-test has that self-test run by CI.
Derived gate union re-run on the final head (dispatch-gates no-path mode, banner names commit a3d95518): check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:entry-guard, check:parse-guard, check:pm-dispatch-gates, check:pnpm-filter-targets, check:watch-hint-literal, check-ci-filter-parity, check-self-test-wired, plus check:nul-bytes — all exit 0, each verdict read from the gate's own judgment line.
NOT run locally (declared narrowing): check-shard-attestation.mjs, check-test-completeness.mjs, check-cross-package-test-inputs.mjs in their ci.yml forms — they read CI run artifacts (test logs) that do not exist locally; CI runs them on this PR.

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

hintCovers does not follow a final-segment basename glob: .changeset/*.md covers zero tracked files while the tree holds ~400 matching changesets, so several changeset gates' hints measure dead and the reason renderer calls it layout moved, which is false. Verdict-moving to fix; not this card.
check:doc-anchors carries the fabricated hint scripts/scripts/check-doc-anchors.mjs: line 459 of that gate spells ./scripts/check-doc-anchors.mjs inside a remedy-message string meant to run from the repo root, and the module-relative resolve doubles the prefix. Now annotated; the spelling fix belongs to that gate.

This PR is skip-changeset: the whole diff is scripts/pm/dispatch-gates.mjs, internal PM tooling — nothing is published from a package.

Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

🤖 Generated with Claude Code

Generated by Claude Code

…fuse @-scoped specifiers as paths Three of the five hints --residue showed for check:query-options-erasure named files that have never existed, printed verbatim: the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled. Measured across all families before choosing the route: 65 of 151 declaring families carried 598 such literals, 353 of them @-scoped package specifiers. - deadHintSweep: the reachability sweep at per-hint grain; the unreachable listing and the new residue annotations read one sweep, two grains. - residueNames / deadNamesNote: dead literals are marked where shown and counted with reasons where not. - isNonPathNamespace: an @-headed first segment (npm scope, version-suffixed specifier, bundler alias) is a namespace, not a path — 353 dead hints refused at extraction, zero tracked paths start with @, no live hint moves. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

os-project-manager · 2026-08-30T12:27:18Z

Fixes#13312

The defect, and the measurement that chose the route

dispatch-gates --residue printed three literals in check:query-options-erasure's names: line that name files which have never existed in this tree, verbatim, with no dead annotation — the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled.

The card declared its scope unmeasured and listed three routes (extractor / annotator / gate-rename) without deciding. Per the dispatch ruling I re-derived the population across ALL families first, on this branch's base 5f0a9c4ad:

reading	before	after (this PR, head `a3d95518`)
families discovered	187	187
families declaring a population	151	149 (2 moved to honest `undetermined`)
reachable families carrying dead literals	65	49
dead literals riding unannotated in reachable families	598	0 (245 remain, every one marked and counted)
fully-dead families (annotated before and after)	3	1

Shape split of the 598: 353 were @-scoped package specifiers (@objectstack/spec, @fx/canary, @typescript-eslint/parser, @/lib/i18n) — the widest single species, exactly the one the third-instance comment on the card measured. 245 are path-shaped fixture names, owner/repo slugs, action specs and drifted doc examples — no closed shape can refuse those, so they are annotated instead of removed.

The measurement picked the route: gate-rename was dead on arrival (it fixes 3 of 598), and the numbers chose one narrow extractor refusal plus the unconditional annotator.

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Annotator (unconditional per triage).deadHintSweep is the existing reachability sweep at per-hint grain; unreachableFamilies now reads its answer off it (one sweep, two grains, no disagreement — pinned). Under --residue, a dead literal is marked where shown and counted where not:

- pnpm check:query-options-erasure [lint.yml] names: scripts/query-options-erasure-baseline.json, scripts/__closure_fixture_gate__.mjs X, scripts/__closure_fixture_helper__.mjs X, ...
3 of 8 declared literal(s) reach nothing tracked (marked X where shown) - dead leads, not population: ...

(The live output uses the ballot-X mark; spelled as X here to keep this body free of markup-sensitive characters.) Display only — no verdict moves; matched/undetermined/silent and the unreachable listing are byte-compatible except for the new marks and notes.

Extractor: isNonPathNamespace gains a third closed shape. A first segment beginning with @ is a scope marker, not a directory — npm grammar, not a heuristic. Zero tracked paths have an @-headed first segment (measured at 5f0a9c4ad), so all 353 refused literals were dead: no live hint moves, no card's verdict moves. The two families whose ENTIRE population was a package-name ledger (check:driver-memory-census, check-test-completeness.mjs — the third-instance comment's table) leave the unreachable listing for undetermined, the bucket that predicate's docblock calls strictly more honest. check:stall-guard-headroom's owner/repo slug is NOT refusable by shape (two bare words are what a path looks like — pinned in that direction) and is now annotated instead.

Zone 2 hypothesis: falsified, with numbers

The PM's hypothesis was that the extractor should stop scraping self-test fixture constants because "a fixture-name constant is not a declaration in any sense." I prototyped exactly that (mask module-scope constants whose identifier is referenced only inside self-test bodies, iterated to a fixed point) and measured it: 96 hints removed — 10 dead, 86 LIVE. The live removals include packages/** and skills/** subtree declarations, whole published-files rosters, and baseline artifacts — because this repo's sanctioned declaration idiom (ROOT_DIR_WATCH_HINTS: a module-scope constant that exists FOR the scanner, referenced only by the self-test that pins it) is structurally indistinguishable from a fixture constant. Reference analysis deletes 8.6 real populations per fabrication removed. In this codebase a scanner-facing constant IS a declaration, so the premise fails and that route is refused. The fixture-name class stays in the hint sets and is now annotated.

The three non-empty delivery controls (from triage)

A real hint still prints unmarked: scripts/query-options-erasure-baseline.json leads the family's names: line above; pinned as ...and leaves the real hint beside it unmarked.
A planted dead literal in a reachable family IS annotated: pinned — scripts/__planted_never_existed__.mjs planted beside a live hint is swept, marked in the names fragment, and counted 1 of 2 with its reason in the note.
Population numbers: table above; 65 of 151 declaring families carried 598 unannotated dead literals, after the fix 0 unannotated (353 refused at extraction, 245 annotated).

Verification (all at head `a3d95518`, exit codes captured before any pipe)

pnpm check:pm-dispatch-gates — verdict line: dispatch-gates self-test: 944 cases pass. (includes the ~30 new pins; one pre-existing pin's specimen-finder needed a third exclusion — the job-filter key, #12956's channel — because thinning the inherited pairs made its .find land on a job-filtered family first; the pin's claim is unchanged)
node scripts/pm/bare-root-worklist.mjs --self-test — OK self-test: 51 live row(s), 43 unreachable as spelled, 43 recorded verdict(s) - none stale, none missing, none contradicted.
node scripts/check-self-test-wired.mjs — every one of the 151 script(s) CI runs that ship a --self-test has that self-test run by CI.
Derived gate union re-run on the final head (dispatch-gates no-path mode, banner names commit a3d95518): check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:entry-guard, check:parse-guard, check:pm-dispatch-gates, check:pnpm-filter-targets, check:watch-hint-literal, check-ci-filter-parity, check-self-test-wired, plus check:nul-bytes — all exit 0, each verdict read from the gate's own judgment line.
NOT run locally (declared narrowing): check-shard-attestation.mjs, check-test-completeness.mjs, check-cross-package-test-inputs.mjs in their ci.yml forms — they read CI run artifacts (test logs) that do not exist locally; CI runs them on this PR.

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

hintCovers does not follow a final-segment basename glob: .changeset/*.md covers zero tracked files while the tree holds ~400 matching changesets, so several changeset gates' hints measure dead and the reason renderer calls it layout moved, which is false. Verdict-moving to fix; not this card.
check:doc-anchors carries the fabricated hint scripts/scripts/check-doc-anchors.mjs: line 459 of that gate spells ./scripts/check-doc-anchors.mjs inside a remedy-message string meant to run from the repo root, and the module-relative resolve doubles the prefix. Now annotated; the spelling fix belongs to that gate.

This PR is skip-changeset: the whole diff is scripts/pm/dispatch-gates.mjs, internal PM tooling — nothing is published from a package.

Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

🤖 Generated with Claude Code

Generated by Claude Code

…fuse @-scoped specifiers as paths Three of the five hints --residue showed for check:query-options-erasure named files that have never existed, printed verbatim: the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled. Measured across all families before choosing the route: 65 of 151 declaring families carried 598 such literals, 353 of them @-scoped package specifiers. - deadHintSweep: the reachability sweep at per-hint grain; the unreachable listing and the new residue annotations read one sweep, two grains. - residueNames / deadNamesNote: dead literals are marked where shown and counted with reasons where not. - isNonPathNamespace: an @-headed first segment (npm scope, version-suffixed specifier, bundler alias) is a namespace, not a path — 353 dead hints refused at extraction, zero tracked paths start with @, no live hint moves. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

os-project-manager · 2026-08-30T12:27:18Z

Fixes#13312

The defect, and the measurement that chose the route

dispatch-gates --residue printed three literals in check:query-options-erasure's names: line that name files which have never existed in this tree, verbatim, with no dead annotation — the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled.

The card declared its scope unmeasured and listed three routes (extractor / annotator / gate-rename) without deciding. Per the dispatch ruling I re-derived the population across ALL families first, on this branch's base 5f0a9c4ad:

reading	before	after (this PR, head `a3d95518`)
families discovered	187	187
families declaring a population	151	149 (2 moved to honest `undetermined`)
reachable families carrying dead literals	65	49
dead literals riding unannotated in reachable families	598	0 (245 remain, every one marked and counted)
fully-dead families (annotated before and after)	3	1

Shape split of the 598: 353 were @-scoped package specifiers (@objectstack/spec, @fx/canary, @typescript-eslint/parser, @/lib/i18n) — the widest single species, exactly the one the third-instance comment on the card measured. 245 are path-shaped fixture names, owner/repo slugs, action specs and drifted doc examples — no closed shape can refuse those, so they are annotated instead of removed.

The measurement picked the route: gate-rename was dead on arrival (it fixes 3 of 598), and the numbers chose one narrow extractor refusal plus the unconditional annotator.

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Annotator (unconditional per triage).deadHintSweep is the existing reachability sweep at per-hint grain; unreachableFamilies now reads its answer off it (one sweep, two grains, no disagreement — pinned). Under --residue, a dead literal is marked where shown and counted where not:

- pnpm check:query-options-erasure [lint.yml] names: scripts/query-options-erasure-baseline.json, scripts/__closure_fixture_gate__.mjs X, scripts/__closure_fixture_helper__.mjs X, ...
3 of 8 declared literal(s) reach nothing tracked (marked X where shown) - dead leads, not population: ...

(The live output uses the ballot-X mark; spelled as X here to keep this body free of markup-sensitive characters.) Display only — no verdict moves; matched/undetermined/silent and the unreachable listing are byte-compatible except for the new marks and notes.

Extractor: isNonPathNamespace gains a third closed shape. A first segment beginning with @ is a scope marker, not a directory — npm grammar, not a heuristic. Zero tracked paths have an @-headed first segment (measured at 5f0a9c4ad), so all 353 refused literals were dead: no live hint moves, no card's verdict moves. The two families whose ENTIRE population was a package-name ledger (check:driver-memory-census, check-test-completeness.mjs — the third-instance comment's table) leave the unreachable listing for undetermined, the bucket that predicate's docblock calls strictly more honest. check:stall-guard-headroom's owner/repo slug is NOT refusable by shape (two bare words are what a path looks like — pinned in that direction) and is now annotated instead.

Zone 2 hypothesis: falsified, with numbers

The PM's hypothesis was that the extractor should stop scraping self-test fixture constants because "a fixture-name constant is not a declaration in any sense." I prototyped exactly that (mask module-scope constants whose identifier is referenced only inside self-test bodies, iterated to a fixed point) and measured it: 96 hints removed — 10 dead, 86 LIVE. The live removals include packages/** and skills/** subtree declarations, whole published-files rosters, and baseline artifacts — because this repo's sanctioned declaration idiom (ROOT_DIR_WATCH_HINTS: a module-scope constant that exists FOR the scanner, referenced only by the self-test that pins it) is structurally indistinguishable from a fixture constant. Reference analysis deletes 8.6 real populations per fabrication removed. In this codebase a scanner-facing constant IS a declaration, so the premise fails and that route is refused. The fixture-name class stays in the hint sets and is now annotated.

The three non-empty delivery controls (from triage)

A real hint still prints unmarked: scripts/query-options-erasure-baseline.json leads the family's names: line above; pinned as ...and leaves the real hint beside it unmarked.
A planted dead literal in a reachable family IS annotated: pinned — scripts/__planted_never_existed__.mjs planted beside a live hint is swept, marked in the names fragment, and counted 1 of 2 with its reason in the note.
Population numbers: table above; 65 of 151 declaring families carried 598 unannotated dead literals, after the fix 0 unannotated (353 refused at extraction, 245 annotated).

Verification (all at head `a3d95518`, exit codes captured before any pipe)

pnpm check:pm-dispatch-gates — verdict line: dispatch-gates self-test: 944 cases pass. (includes the ~30 new pins; one pre-existing pin's specimen-finder needed a third exclusion — the job-filter key, #12956's channel — because thinning the inherited pairs made its .find land on a job-filtered family first; the pin's claim is unchanged)
node scripts/pm/bare-root-worklist.mjs --self-test — OK self-test: 51 live row(s), 43 unreachable as spelled, 43 recorded verdict(s) - none stale, none missing, none contradicted.
node scripts/check-self-test-wired.mjs — every one of the 151 script(s) CI runs that ship a --self-test has that self-test run by CI.
Derived gate union re-run on the final head (dispatch-gates no-path mode, banner names commit a3d95518): check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:entry-guard, check:parse-guard, check:pm-dispatch-gates, check:pnpm-filter-targets, check:watch-hint-literal, check-ci-filter-parity, check-self-test-wired, plus check:nul-bytes — all exit 0, each verdict read from the gate's own judgment line.
NOT run locally (declared narrowing): check-shard-attestation.mjs, check-test-completeness.mjs, check-cross-package-test-inputs.mjs in their ci.yml forms — they read CI run artifacts (test logs) that do not exist locally; CI runs them on this PR.

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

hintCovers does not follow a final-segment basename glob: .changeset/*.md covers zero tracked files while the tree holds ~400 matching changesets, so several changeset gates' hints measure dead and the reason renderer calls it layout moved, which is false. Verdict-moving to fix; not this card.
check:doc-anchors carries the fabricated hint scripts/scripts/check-doc-anchors.mjs: line 459 of that gate spells ./scripts/check-doc-anchors.mjs inside a remedy-message string meant to run from the repo root, and the module-relative resolve doubles the prefix. Now annotated; the spelling fix belongs to that gate.

This PR is skip-changeset: the whole diff is scripts/pm/dispatch-gates.mjs, internal PM tooling — nothing is published from a package.

Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

🤖 Generated with Claude Code

Generated by Claude Code

…fuse @-scoped specifiers as paths Three of the five hints --residue showed for check:query-options-erasure named files that have never existed, printed verbatim: the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled. Measured across all families before choosing the route: 65 of 151 declaring families carried 598 such literals, 353 of them @-scoped package specifiers. - deadHintSweep: the reachability sweep at per-hint grain; the unreachable listing and the new residue annotations read one sweep, two grains. - residueNames / deadNamesNote: dead literals are marked where shown and counted with reasons where not. - isNonPathNamespace: an @-headed first segment (npm scope, version-suffixed specifier, bundler alias) is a namespace, not a path — 353 dead hints refused at extraction, zero tracked paths start with @, no live hint moves. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

os-project-manager · 2026-08-30T12:27:18Z

Fixes#13312

The defect, and the measurement that chose the route

dispatch-gates --residue printed three literals in check:query-options-erasure's names: line that name files which have never existed in this tree, verbatim, with no dead annotation — the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled.

The card declared its scope unmeasured and listed three routes (extractor / annotator / gate-rename) without deciding. Per the dispatch ruling I re-derived the population across ALL families first, on this branch's base 5f0a9c4ad:

reading	before	after (this PR, head `a3d95518`)
families discovered	187	187
families declaring a population	151	149 (2 moved to honest `undetermined`)
reachable families carrying dead literals	65	49
dead literals riding unannotated in reachable families	598	0 (245 remain, every one marked and counted)
fully-dead families (annotated before and after)	3	1

Shape split of the 598: 353 were @-scoped package specifiers (@objectstack/spec, @fx/canary, @typescript-eslint/parser, @/lib/i18n) — the widest single species, exactly the one the third-instance comment on the card measured. 245 are path-shaped fixture names, owner/repo slugs, action specs and drifted doc examples — no closed shape can refuse those, so they are annotated instead of removed.

The measurement picked the route: gate-rename was dead on arrival (it fixes 3 of 598), and the numbers chose one narrow extractor refusal plus the unconditional annotator.

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Annotator (unconditional per triage).deadHintSweep is the existing reachability sweep at per-hint grain; unreachableFamilies now reads its answer off it (one sweep, two grains, no disagreement — pinned). Under --residue, a dead literal is marked where shown and counted where not:

- pnpm check:query-options-erasure [lint.yml] names: scripts/query-options-erasure-baseline.json, scripts/__closure_fixture_gate__.mjs X, scripts/__closure_fixture_helper__.mjs X, ...
3 of 8 declared literal(s) reach nothing tracked (marked X where shown) - dead leads, not population: ...

(The live output uses the ballot-X mark; spelled as X here to keep this body free of markup-sensitive characters.) Display only — no verdict moves; matched/undetermined/silent and the unreachable listing are byte-compatible except for the new marks and notes.

Extractor: isNonPathNamespace gains a third closed shape. A first segment beginning with @ is a scope marker, not a directory — npm grammar, not a heuristic. Zero tracked paths have an @-headed first segment (measured at 5f0a9c4ad), so all 353 refused literals were dead: no live hint moves, no card's verdict moves. The two families whose ENTIRE population was a package-name ledger (check:driver-memory-census, check-test-completeness.mjs — the third-instance comment's table) leave the unreachable listing for undetermined, the bucket that predicate's docblock calls strictly more honest. check:stall-guard-headroom's owner/repo slug is NOT refusable by shape (two bare words are what a path looks like — pinned in that direction) and is now annotated instead.

Zone 2 hypothesis: falsified, with numbers

The PM's hypothesis was that the extractor should stop scraping self-test fixture constants because "a fixture-name constant is not a declaration in any sense." I prototyped exactly that (mask module-scope constants whose identifier is referenced only inside self-test bodies, iterated to a fixed point) and measured it: 96 hints removed — 10 dead, 86 LIVE. The live removals include packages/** and skills/** subtree declarations, whole published-files rosters, and baseline artifacts — because this repo's sanctioned declaration idiom (ROOT_DIR_WATCH_HINTS: a module-scope constant that exists FOR the scanner, referenced only by the self-test that pins it) is structurally indistinguishable from a fixture constant. Reference analysis deletes 8.6 real populations per fabrication removed. In this codebase a scanner-facing constant IS a declaration, so the premise fails and that route is refused. The fixture-name class stays in the hint sets and is now annotated.

The three non-empty delivery controls (from triage)

A real hint still prints unmarked: scripts/query-options-erasure-baseline.json leads the family's names: line above; pinned as ...and leaves the real hint beside it unmarked.
A planted dead literal in a reachable family IS annotated: pinned — scripts/__planted_never_existed__.mjs planted beside a live hint is swept, marked in the names fragment, and counted 1 of 2 with its reason in the note.
Population numbers: table above; 65 of 151 declaring families carried 598 unannotated dead literals, after the fix 0 unannotated (353 refused at extraction, 245 annotated).

Verification (all at head `a3d95518`, exit codes captured before any pipe)

pnpm check:pm-dispatch-gates — verdict line: dispatch-gates self-test: 944 cases pass. (includes the ~30 new pins; one pre-existing pin's specimen-finder needed a third exclusion — the job-filter key, #12956's channel — because thinning the inherited pairs made its .find land on a job-filtered family first; the pin's claim is unchanged)
node scripts/pm/bare-root-worklist.mjs --self-test — OK self-test: 51 live row(s), 43 unreachable as spelled, 43 recorded verdict(s) - none stale, none missing, none contradicted.
node scripts/check-self-test-wired.mjs — every one of the 151 script(s) CI runs that ship a --self-test has that self-test run by CI.
Derived gate union re-run on the final head (dispatch-gates no-path mode, banner names commit a3d95518): check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:cross-package-test-inputs, check:entry-guard, check:parse-guard, check:pm-dispatch-gates, check:pnpm-filter-targets, check:watch-hint-literal, check-ci-filter-parity, check-self-test-wired, plus check:nul-bytes — all exit 0, each verdict read from the gate's own judgment line.
NOT run locally (declared narrowing): check-shard-attestation.mjs, check-test-completeness.mjs, check-cross-package-test-inputs.mjs in their ci.yml forms — they read CI run artifacts (test logs) that do not exist locally; CI runs them on this PR.

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

hintCovers does not follow a final-segment basename glob: .changeset/*.md covers zero tracked files while the tree holds ~400 matching changesets, so several changeset gates' hints measure dead and the reason renderer calls it layout moved, which is false. Verdict-moving to fix; not this card.
check:doc-anchors carries the fabricated hint scripts/scripts/check-doc-anchors.mjs: line 459 of that gate spells ./scripts/check-doc-anchors.mjs inside a remedy-message string meant to run from the repo root, and the module-relative resolve doubles the prefix. Now annotated; the spelling fix belongs to that gate.

This PR is skip-changeset: the whole diff is scripts/pm/dispatch-gates.mjs, internal PM tooling — nothing is published from a package.

Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

🤖 Generated with Claude Code

Generated by Claude Code

…fuse @-scoped specifiers as paths Three of the five hints --residue showed for check:query-options-erasure named files that have never existed, printed verbatim: the annotator only spoke when a WHOLE family was unreachable, so one live baseline walked its dead siblings past every reader unlabelled. Measured across all families before choosing the route: 65 of 151 declaring families carried 598 such literals, 353 of them @-scoped package specifiers. - deadHintSweep: the reachability sweep at per-hint grain; the unreachable listing and the new residue annotations read one sweep, two grains. - residueNames / deadNamesNote: dead literals are marked where shown and counted with reasons where not. - isNonPathNamespace: an @-headed first segment (npm scope, version-suffixed specifier, bundler alias) is a namespace, not a path — 353 dead hints refused at extraction, zero tracked paths start with @, no live hint moves. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pk26oZ12t5N1hwGW1m1MgC

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Aug 30, 2026

github-actionsBot added the size/m label Aug 30, 2026

os-project-manager marked this pull request as ready for review August 30, 2026 12:43

os-project-manager enabled auto-merge August 30, 2026 12:43

os-project-manager added this pull request to the merge queue Aug 30, 2026

Merged via the queue into main with commit 7927f4cAug 30, 2026
34 checks passed

os-project-manager deleted the claude/issue-13312-residue-fabricated-leads branch August 30, 2026 13:11

os-project-manager mentioned this pull request Aug 30, 2026
[PM seat] domain:devx @ objectstack — 🔵 已收班 R34 · session_01Pk26oZ12t5N1hwGW1m1MgC · 落地 7 卡 / 7 PR · 立卡 11 · 零返工零踢出 · ⛔ 零留守尾巴 · 简报见最新评论 #6023
Open

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Aug 30, 2026

github-actionsBot added the size/m label Aug 30, 2026

os-project-manager marked this pull request as ready for review August 30, 2026 12:43

os-project-manager enabled auto-merge August 30, 2026 12:43

os-project-manager added this pull request to the merge queue Aug 30, 2026

Merged via the queue into main with commit 7927f4cAug 30, 2026
34 checks passed

os-project-manager deleted the claude/issue-13312-residue-fabricated-leads branch August 30, 2026 13:11

os-project-manager mentioned this pull request Aug 30, 2026
[PM seat] domain:devx @ objectstack — 🔵 已收班 R34 · session_01Pk26oZ12t5N1hwGW1m1MgC · 落地 7 卡 / 7 PR · 立卡 11 · 零返工零踢出 · ⛔ 零留守尾巴 · 简报见最新评论 #6023
Open

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Aug 30, 2026

github-actionsBot added the size/m label Aug 30, 2026

os-project-manager marked this pull request as ready for review August 30, 2026 12:43

os-project-manager enabled auto-merge August 30, 2026 12:43

os-project-manager added this pull request to the merge queue Aug 30, 2026

Merged via the queue into main with commit 7927f4cAug 30, 2026
34 checks passed

os-project-manager deleted the claude/issue-13312-residue-fabricated-leads branch August 30, 2026 13:11

os-project-manager mentioned this pull request Aug 30, 2026
[PM seat] domain:devx @ objectstack — 🔵 已收班 R34 · session_01Pk26oZ12t5N1hwGW1m1MgC · 落地 7 卡 / 7 PR · 立卡 11 · 零返工零踢出 · ⛔ 零留守尾巴 · 简报见最新评论 #6023
Open

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Aug 30, 2026

github-actionsBot added the size/m label Aug 30, 2026

os-project-manager marked this pull request as ready for review August 30, 2026 12:43

os-project-manager enabled auto-merge August 30, 2026 12:43

os-project-manager added this pull request to the merge queue Aug 30, 2026

Merged via the queue into main with commit 7927f4cAug 30, 2026
34 checks passed

os-project-manager deleted the claude/issue-13312-residue-fabricated-leads branch August 30, 2026 13:11

os-project-manager mentioned this pull request Aug 30, 2026
[PM seat] domain:devx @ objectstack — 🔵 已收班 R34 · session_01Pk26oZ12t5N1hwGW1m1MgC · 落地 7 卡 / 7 PR · 立卡 11 · 零返工零踢出 · ⛔ 零留守尾巴 · 简报见最新评论 #6023
Open

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Aug 30, 2026

github-actionsBot added the size/m label Aug 30, 2026

os-project-manager marked this pull request as ready for review August 30, 2026 12:43

os-project-manager enabled auto-merge August 30, 2026 12:43

os-project-manager added this pull request to the merge queue Aug 30, 2026

Merged via the queue into main with commit 7927f4cAug 30, 2026
34 checks passed

os-project-manager deleted the claude/issue-13312-residue-fabricated-leads branch August 30, 2026 13:11

os-project-manager mentioned this pull request Aug 30, 2026
[PM seat] domain:devx @ objectstack — 🔵 已收班 R34 · session_01Pk26oZ12t5N1hwGW1m1MgC · 落地 7 卡 / 7 PR · 立卡 11 · 零返工零踢出 · ⛔ 零留守尾巴 · 简报见最新评论 #6023
Open

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Aug 30, 2026

github-actionsBot added the size/m label Aug 30, 2026

os-project-manager marked this pull request as ready for review August 30, 2026 12:43

os-project-manager enabled auto-merge August 30, 2026 12:43

os-project-manager added this pull request to the merge queue Aug 30, 2026

Merged via the queue into main with commit 7927f4cAug 30, 2026
34 checks passed

os-project-manager deleted the claude/issue-13312-residue-fabricated-leads branch August 30, 2026 13:11

os-project-manager mentioned this pull request Aug 30, 2026
[PM seat] domain:devx @ objectstack — 🔵 已收班 R34 · session_01Pk26oZ12t5N1hwGW1m1MgC · 落地 7 卡 / 7 PR · 立卡 11 · 零返工零踢出 · ⛔ 零留守尾巴 · 简报见最新评论 #6023
Open

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Aug 30, 2026

github-actionsBot added the size/m label Aug 30, 2026

os-project-manager marked this pull request as ready for review August 30, 2026 12:43

os-project-manager enabled auto-merge August 30, 2026 12:43

os-project-manager added this pull request to the merge queue Aug 30, 2026

Merged via the queue into main with commit 7927f4cAug 30, 2026
34 checks passed

os-project-manager deleted the claude/issue-13312-residue-fabricated-leads branch August 30, 2026 13:11

os-project-manager mentioned this pull request Aug 30, 2026
[PM seat] domain:devx @ objectstack — 🔵 已收班 R34 · session_01Pk26oZ12t5N1hwGW1m1MgC · 落地 7 卡 / 7 PR · 立卡 11 · 零返工零踢出 · ⛔ 零留守尾巴 · 简报见最新评论 #6023
Open

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Aug 30, 2026

github-actionsBot added the size/m label Aug 30, 2026

os-project-manager marked this pull request as ready for review August 30, 2026 12:43

os-project-manager enabled auto-merge August 30, 2026 12:43

os-project-manager added this pull request to the merge queue Aug 30, 2026

Merged via the queue into main with commit 7927f4cAug 30, 2026
34 checks passed

os-project-manager deleted the claude/issue-13312-residue-fabricated-leads branch August 30, 2026 13:11

os-project-manager mentioned this pull request Aug 30, 2026
[PM seat] domain:devx @ objectstack — 🔵 已收班 R34 · session_01Pk26oZ12t5N1hwGW1m1MgC · 落地 7 卡 / 7 PR · 立卡 11 · 零返工零踢出 · ⛔ 零留守尾巴 · 简报见最新评论 #6023
Open

Uh oh!

Conversation

os-project-manager commented Aug 30, 2026

The defect, and the measurement that chose the route

What changed (one file: scripts/pm/dispatch-gates.mjs)

Zone 2 hypothesis: falsified, with numbers

The three non-empty delivery controls (from triage)

Verification (all at head a3d95518, exit codes captured before any pipe)

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

os-project-manager commented Aug 30, 2026

The defect, and the measurement that chose the route

What changed (one file: scripts/pm/dispatch-gates.mjs)

Zone 2 hypothesis: falsified, with numbers

The three non-empty delivery controls (from triage)

Verification (all at head a3d95518, exit codes captured before any pipe)

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

os-project-manager commented Aug 30, 2026

The defect, and the measurement that chose the route

What changed (one file: scripts/pm/dispatch-gates.mjs)

Zone 2 hypothesis: falsified, with numbers

The three non-empty delivery controls (from triage)

Verification (all at head a3d95518, exit codes captured before any pipe)

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

os-project-manager commented Aug 30, 2026

The defect, and the measurement that chose the route

What changed (one file: scripts/pm/dispatch-gates.mjs)

Zone 2 hypothesis: falsified, with numbers

The three non-empty delivery controls (from triage)

Verification (all at head a3d95518, exit codes captured before any pipe)

Out of scope, surfaced by the new annotation (handed to the PM, dedup channel degraded)

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

os-project-manager commented Aug 30, 2026

The defect, and the measurement that chose the route

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Verification (all at head `a3d95518`, exit codes captured before any pipe)

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Verification (all at head `a3d95518`, exit codes captured before any pipe)

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Verification (all at head `a3d95518`, exit codes captured before any pipe)

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Verification (all at head `a3d95518`, exit codes captured before any pipe)

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Verification (all at head `a3d95518`, exit codes captured before any pipe)

What changed (one file: `scripts/pm/dispatch-gates.mjs`)

Verification (all at head `a3d95518`, exit codes captured before any pipe)