fix(pm): read a package-root-anchored literal against the root its writer binds, and stop the residue calling it a non-path by claude[bot] · Pull Request #14450 · objectstack-ai/objectstack

claude · 2026-09-02T07:27:23Z

Two deliverables in one PR on one file, scripts/pm/dispatch-gates.mjs, as three commits: the package-root anchor (#14208), the residue reason text it is coupled to (#14208), and the read-edge prose repair (#14289, no behaviour change).

#14208 — measure first, as the card required

Premise re-confirmed on this PR's base (ad54eb342), with the card's own command:

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --residue packages/spec/src/index.ts

check:generated printed 5 of 10 declared literal(s) reach nothing tracked, the three named literals among them, each under the reason no tracked path under its first segment; never was a repo path — while all three sit on disk under packages/spec/.

The population sweep

Over the 201 discovered families' 195 scannable gate scripts, resolving every import.meta-seeded binding through the tool's own resolvePathExpression:

measurement	value
discovered gate scripts binding a package root	8 (all `packages/spec`)
...of those, scripts carrying a hint dead at the repo root	2
dead-at-root hints in that population	6
...that re-anchor onto something the tree HAS	5
...refused because the re-anchored form reaches nothing	1 (`'zod'`, a package specifier)

⚠️The card's description of the shape does not survive the sweep, and this changed the rule. The card described the class as literals passed to resolve(PKG_DIR, …). That shape is real — 6 such literals across 6 of the 8 scripts, every one a tracked file — but it is not where the confirmed instance comes from. All five of check:generated's dead literals are artifact: values in its GATED ledger (artifact: 'api-surface/', artifact: 'src/meta-spelling/meta-url-data.generated.ts', …), describing what each gate verifies; nothing resolves them. A rule restricted to the syntactic shape would have repaired none of the three dead leads the card names. The restriction was measured and dropped; the reachability test is what keeps the wider reading honest.

The rule that landed

A hint is re-read against the package root its writer binds only when deepestTrackedPrefix is empty — no tracked path begins with even its first segment, which is exactly the branch whose reason text was false — and only if the re-anchored form reaches the tree through hintReachesTree. It replaces rather than adds. Neither priced refusal is relaxed and neither is reachable from here: admission is untouched (a literal that is not already a hint never arrives), and a re-anchoring that names nothing is dropped, so the rule cannot invent a path. The layout-moved class (a shorter surviving prefix) is deliberately left alone — its first segment is a real repo directory, so re-anchoring it would replace a triage lead with a fabricated one.

The price, before and after, on this base

Over 201 discovered families and 7900 tracked files:

	before	after
watch-hint (gate, file) pairs	142089	142139 (+50, 0 lost)
families whose hint set moves	—	1 (`check:generated`, no other)
pairs re-attributed (same pair, new via)	—	0 — every replaced spelling was dead, so it covered nothing to re-attribute
(check, hint) live / inert	1108 / 152	1113 / 147
distinct hints in the fleet	725	722 — three of the five re-anchored forms are already spelled from the root by other gates
families carrying dead literals	46	45
dead literals	152	147
...in the empty-`deepest` class	128	123

Provenance, which is the criterion this file prices and never volume: 5 of 5 admitted hints are TRUE leads, and 50 of 50 pairs with them, each verified at its declaration site — they are check:generated's own artifact ledger and verifying those artifacts is the gate's whole job:

re-anchored hint	files	ledger entry
`packages/spec/api-surface`	17	`check:api-surface`
`packages/spec/export-origins`	17	`check:export-origins`
`packages/spec/declaration-map`	14	`check:declaration-map`
`packages/spec/src/meta-spelling/meta-url-data.generated.ts`	1	`check:meta-url-spelling`
`packages/spec/liveness/state-counts.md`	1	the liveness counts

0 fabricated. For scale, hintCovers' docblock prices the bare-top-level-word admission it refuses at +139084 pairs on this corpus; this is 0.04% of it, and moduleRelativeDirectoryHint's naive-resolved-form widening was +53 hints for 1 true lead against this rule's 5 for 5.

The whole --residue output moves by exactly one row before and after (diffed line by line).

The reason text

The by-construction sentence ended never was a repo path — a claim about EVERY base made from evidence about ONE, and a dead-lead label printed over three live files. It now reads:

'HINT' — no tracked path under its first segment; a specifier, or a path anchored at a base this scan did not resolve

Checkable by the reader (git ls-files on the first segment is empty) and no longer asserting a base it did not test. The three docblocks stating the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep still lands on the incident. The pins move with the text: pinned to a retired phrase, the dist-freshness case would pass on any tree at all — a green over the exact regression it exists to catch.

Ablation

Anchor call replaced with return [...hints] on the committed tree, mutation proven on disk before the run (anchor occurrences 1 to 0, injected marker 1, blob 9743275c to e51f9d24), restore leg trapped and then proven by byte comparison (git diff HEAD empty, blob back to the HEAD blob 9743275c). Ablated, check:generated returns to 5 of 10 declared literal(s) reach nothing tracked with the three named literals marked. The ablated run also shows the corrected reason text in place, so the two halves are independent — the reason-text repair is the deliverable the card specified for the no-repair branch, and it stands on its own.

#14289 — prose only, no behaviour change

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM … and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the last key in coveringKey, an identity claim about a single FILE (#13000).

Re-derived at this PR's base, not copied from the card:

EDGE import: 27 families, 38 (family,target) pairs, 15 distinct targets
EDGE run: 1 family, 1 pair, 1 distinct target
EDGE manifest: 10 families, 10 pairs, 2 distinct targets
EDGE read: 12 families, 16 pairs, 15 distinct targets
INHERITED HINTS BY EDGE LABEL:
import: 373 hints across 13 families
manifest: 7 hints across 7 families
run: 1 hint across 1 family
read: 0

Read targets that carried a hint into their family: 0. Of the 16 read-edge pairs, 8 are answered by the READ identity key and 8 by an earlier key — consistent with coveringKey's "additive by construction" note.

The two numberings are reconciled rather than flattened: the docblock numbers all four scans in source order (so the manifest edge is the fourth), discoverFamilies numbers only the population follows (so the same edge is its third), and each now says which. The claim that there are three population follows all ending at a program is gone, and packageManifestTargets' "the other three follows take" is corrected to two.

Gates

Union re-run after the final commit, at git rev-parse --short HEAD = 70c1aedff (merge of origin/maind16df7418). Derived with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands — 15 families, all named by path, 0 by kind; the change set is exactly the one file. Exit codes captured to a file before any pipe.

gate	exit	verdict
`pnpm check:pm-dispatch-gates` (the tool's `--self-test`)	0	RAN-PASS — `dispatch-gates self-test: 1240 cases pass`
`node scripts/check-ci-filter-parity.mjs`	0	RAN-PASS
`node scripts/check-cross-package-test-inputs.mjs` / `pnpm check:cross-package-test-inputs`	0 / 0	RAN-PASS (both spellings)
`node scripts/check-self-test-wired.mjs`	0	RAN-PASS
`node scripts/check-shard-attestation.mjs`	0	RAN-PASS
`node scripts/check-test-completeness.mjs`	3	NOT MEASURED — no test log to hand it; the gate says so itself and calls exit 3 distinct from a finding's 1
`pnpm check:agent-test-spelling`	0	RAN-PASS
`pnpm check:bash32-floor`	0	RAN-PASS
`pnpm check:cli-command-ids`	0	RAN-PASS
`pnpm check:declared-population-live`	0	RAN-PASS
`pnpm check:entry-guard`	0	RAN-PASS
`pnpm check:parse-guard`	0	RAN-PASS
`pnpm check:pnpm-filter-targets`	0	RAN-PASS
`pnpm check:watch-hint-literal`	0	RAN-PASS
`pnpm check:ratchet-remedy-authority`	0	RAN-PASS (beyond the derived family)
`node scripts/check-self-test-workflow-commands.mjs`	0	RAN-PASS (beyond the derived family)
`pnpm check:nul-bytes`	0	RAN-PASS

Lint, narrowed and declared.pnpm lint is eslint . --no-inline-config over the whole repo; the run here is targeted at the one changed file: 1 file linted, 0 errors, 0 warnings, exit 0, the count read from --format json. The narrowing is a measurement rather than a gap because this repo runs one eslint.config.mjs which never enables type-aware linting for any file (no parserOptions.project, no typed rules — stated and positive-control-measured in that config's own header), so a one-file diff cannot move the verdict on any untouched file, and the file is a config match rather than an ignore (it produced a result object, not an ignore warning).

Changeset: none. This is a scripts/**-only diff that publishes nothing from any package, which is the repo convention on recent scripts-only PRs; the skip-changeset label carries it.

Generated by Claude Code

…iter binds extractWatchHints resolves a path literal against exactly two anchors: the repo root, and the writer's own directory when the author spelled a ./ or ../ prefix. A gate that lives inside a package binds a third at module top -- join(dirname(fileURLToPath(import.meta.url)), '..') -- and every literal it writes against that binding was read from the REPO root, where nothing of the sort exists. check:generated's five artifact-ledger literals are the confirmed instance: all five alive under packages/spec/ and all five printed by the residue as dead leads, under a reason text asserting they were never repo paths. Measured first, as the card required. Over the 201 discovered families' 195 scannable gate scripts, 8 bind a package root this way, 2 of them carry a hint dead at the repo root, 6 such hints exist, 5 re-anchor onto something the tree has and 1 is refused because it reaches nothing. Price: 142089 -> 142139 watch-hint (gate, file) pairs (+50, 0 lost, 0 re-attributed), one family's hint set moves, and 5 of 5 admitted hints are true leads verified at their declaration site. The admitted class is exactly the one whose residue sentence was false: a hint is re-anchored only when no tracked path begins with even its first segment. The layout-moved class is left to triage, admission is untouched, and a re-anchoring that reaches nothing is dropped, so the rule cannot invent a path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

unreachableReason's by-construction branch ended "never was a repo path" -- a claim about EVERY base, made from evidence about ONE. check:generated's api-surface, export-origins and src/meta-spelling/... all printed under it while sitting on disk under packages/spec/: a dead-lead label on a live path, telling the reader there is nothing to chase about three files they could open. The sentence now names what the sweep actually established (no tracked path under the literal's first segment) and the two ways that happens, without asserting the literal never existed anywhere. The three docblocks that stated the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep for it still lands on the incident. The pins move with the text rather than staying on the retired phrase: pinned to a string nothing renders, the dist-freshness case would pass on any tree at all -- a green over the exact regression it exists to catch -- so it now asserts against the sentence that branch prints today, and the sweep case asserts both halves. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM ... and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the LAST key in coveringKey, which is an identity claim about a single FILE and never a population. Re-derived at this PR's base: the read edge is 12 families, 16 (family,target) pairs over 15 distinct targets, and it carries 0 inherited hints -- against 373 for the import edge, 7 for the manifest edge and 1 for the run edge. Reading a program's TEXT is not performing its work, so identity-only is the correct behaviour and the sentence was the defect. No behaviour change. The two numberings inside the file also disagreed: discoverFamilies called the spawn follow the SECOND edge and the manifest follow the THIRD, while the docblock called the manifest one the FOURTH. They count different things -- source-order edges, and population follows -- so each now says which, and the claim that there are three population follows all ending at a program is gone. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

…spatch-gates-package-root-anchor

claude · 2026-09-02T07:27:23Z

Fixes#14208
Fixes#14289

Two deliverables in one PR on one file, scripts/pm/dispatch-gates.mjs, as three commits: the package-root anchor (#14208), the residue reason text it is coupled to (#14208), and the read-edge prose repair (#14289, no behaviour change).

#14208 — measure first, as the card required

Premise re-confirmed on this PR's base (ad54eb342), with the card's own command:

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --residue packages/spec/src/index.ts

check:generated printed 5 of 10 declared literal(s) reach nothing tracked, the three named literals among them, each under the reason no tracked path under its first segment; never was a repo path — while all three sit on disk under packages/spec/.

The population sweep

Over the 201 discovered families' 195 scannable gate scripts, resolving every import.meta-seeded binding through the tool's own resolvePathExpression:

measurement	value
discovered gate scripts binding a package root	8 (all `packages/spec`)
...of those, scripts carrying a hint dead at the repo root	2
dead-at-root hints in that population	6
...that re-anchor onto something the tree HAS	5
...refused because the re-anchored form reaches nothing	1 (`'zod'`, a package specifier)

⚠️The card's description of the shape does not survive the sweep, and this changed the rule. The card described the class as literals passed to resolve(PKG_DIR, …). That shape is real — 6 such literals across 6 of the 8 scripts, every one a tracked file — but it is not where the confirmed instance comes from. All five of check:generated's dead literals are artifact: values in its GATED ledger (artifact: 'api-surface/', artifact: 'src/meta-spelling/meta-url-data.generated.ts', …), describing what each gate verifies; nothing resolves them. A rule restricted to the syntactic shape would have repaired none of the three dead leads the card names. The restriction was measured and dropped; the reachability test is what keeps the wider reading honest.

The rule that landed

A hint is re-read against the package root its writer binds only when deepestTrackedPrefix is empty — no tracked path begins with even its first segment, which is exactly the branch whose reason text was false — and only if the re-anchored form reaches the tree through hintReachesTree. It replaces rather than adds. Neither priced refusal is relaxed and neither is reachable from here: admission is untouched (a literal that is not already a hint never arrives), and a re-anchoring that names nothing is dropped, so the rule cannot invent a path. The layout-moved class (a shorter surviving prefix) is deliberately left alone — its first segment is a real repo directory, so re-anchoring it would replace a triage lead with a fabricated one.

The price, before and after, on this base

Over 201 discovered families and 7900 tracked files:

	before	after
watch-hint (gate, file) pairs	142089	142139 (+50, 0 lost)
families whose hint set moves	—	1 (`check:generated`, no other)
pairs re-attributed (same pair, new via)	—	0 — every replaced spelling was dead, so it covered nothing to re-attribute
(check, hint) live / inert	1108 / 152	1113 / 147
distinct hints in the fleet	725	722 — three of the five re-anchored forms are already spelled from the root by other gates
families carrying dead literals	46	45
dead literals	152	147
...in the empty-`deepest` class	128	123

Provenance, which is the criterion this file prices and never volume: 5 of 5 admitted hints are TRUE leads, and 50 of 50 pairs with them, each verified at its declaration site — they are check:generated's own artifact ledger and verifying those artifacts is the gate's whole job:

re-anchored hint	files	ledger entry
`packages/spec/api-surface`	17	`check:api-surface`
`packages/spec/export-origins`	17	`check:export-origins`
`packages/spec/declaration-map`	14	`check:declaration-map`
`packages/spec/src/meta-spelling/meta-url-data.generated.ts`	1	`check:meta-url-spelling`
`packages/spec/liveness/state-counts.md`	1	the liveness counts

0 fabricated. For scale, hintCovers' docblock prices the bare-top-level-word admission it refuses at +139084 pairs on this corpus; this is 0.04% of it, and moduleRelativeDirectoryHint's naive-resolved-form widening was +53 hints for 1 true lead against this rule's 5 for 5.

The whole --residue output moves by exactly one row before and after (diffed line by line).

The reason text

The by-construction sentence ended never was a repo path — a claim about EVERY base made from evidence about ONE, and a dead-lead label printed over three live files. It now reads:

'HINT' — no tracked path under its first segment; a specifier, or a path anchored at a base this scan did not resolve

Checkable by the reader (git ls-files on the first segment is empty) and no longer asserting a base it did not test. The three docblocks stating the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep still lands on the incident. The pins move with the text: pinned to a retired phrase, the dist-freshness case would pass on any tree at all — a green over the exact regression it exists to catch.

Ablation

Anchor call replaced with return [...hints] on the committed tree, mutation proven on disk before the run (anchor occurrences 1 to 0, injected marker 1, blob 9743275c to e51f9d24), restore leg trapped and then proven by byte comparison (git diff HEAD empty, blob back to the HEAD blob 9743275c). Ablated, check:generated returns to 5 of 10 declared literal(s) reach nothing tracked with the three named literals marked. The ablated run also shows the corrected reason text in place, so the two halves are independent — the reason-text repair is the deliverable the card specified for the no-repair branch, and it stands on its own.

#14289 — prose only, no behaviour change

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM … and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the last key in coveringKey, an identity claim about a single FILE (#13000).

Re-derived at this PR's base, not copied from the card:

EDGE import: 27 families, 38 (family,target) pairs, 15 distinct targets
EDGE run: 1 family, 1 pair, 1 distinct target
EDGE manifest: 10 families, 10 pairs, 2 distinct targets
EDGE read: 12 families, 16 pairs, 15 distinct targets
INHERITED HINTS BY EDGE LABEL:
import: 373 hints across 13 families
manifest: 7 hints across 7 families
run: 1 hint across 1 family
read: 0

Read targets that carried a hint into their family: 0. Of the 16 read-edge pairs, 8 are answered by the READ identity key and 8 by an earlier key — consistent with coveringKey's "additive by construction" note.

The two numberings are reconciled rather than flattened: the docblock numbers all four scans in source order (so the manifest edge is the fourth), discoverFamilies numbers only the population follows (so the same edge is its third), and each now says which. The claim that there are three population follows all ending at a program is gone, and packageManifestTargets' "the other three follows take" is corrected to two.

Gates

Union re-run after the final commit, at git rev-parse --short HEAD = 70c1aedff (merge of origin/maind16df7418). Derived with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands — 15 families, all named by path, 0 by kind; the change set is exactly the one file. Exit codes captured to a file before any pipe.

gate	exit	verdict
`pnpm check:pm-dispatch-gates` (the tool's `--self-test`)	0	RAN-PASS — `dispatch-gates self-test: 1240 cases pass`
`node scripts/check-ci-filter-parity.mjs`	0	RAN-PASS
`node scripts/check-cross-package-test-inputs.mjs` / `pnpm check:cross-package-test-inputs`	0 / 0	RAN-PASS (both spellings)
`node scripts/check-self-test-wired.mjs`	0	RAN-PASS
`node scripts/check-shard-attestation.mjs`	0	RAN-PASS
`node scripts/check-test-completeness.mjs`	3	NOT MEASURED — no test log to hand it; the gate says so itself and calls exit 3 distinct from a finding's 1
`pnpm check:agent-test-spelling`	0	RAN-PASS
`pnpm check:bash32-floor`	0	RAN-PASS
`pnpm check:cli-command-ids`	0	RAN-PASS
`pnpm check:declared-population-live`	0	RAN-PASS
`pnpm check:entry-guard`	0	RAN-PASS
`pnpm check:parse-guard`	0	RAN-PASS
`pnpm check:pnpm-filter-targets`	0	RAN-PASS
`pnpm check:watch-hint-literal`	0	RAN-PASS
`pnpm check:ratchet-remedy-authority`	0	RAN-PASS (beyond the derived family)
`node scripts/check-self-test-workflow-commands.mjs`	0	RAN-PASS (beyond the derived family)
`pnpm check:nul-bytes`	0	RAN-PASS

Lint, narrowed and declared.pnpm lint is eslint . --no-inline-config over the whole repo; the run here is targeted at the one changed file: 1 file linted, 0 errors, 0 warnings, exit 0, the count read from --format json. The narrowing is a measurement rather than a gap because this repo runs one eslint.config.mjs which never enables type-aware linting for any file (no parserOptions.project, no typed rules — stated and positive-control-measured in that config's own header), so a one-file diff cannot move the verdict on any untouched file, and the file is a config match rather than an ignore (it produced a result object, not an ignore warning).

Changeset: none. This is a scripts/**-only diff that publishes nothing from any package, which is the repo convention on recent scripts-only PRs; the skip-changeset label carries it.

Generated by Claude Code

…iter binds extractWatchHints resolves a path literal against exactly two anchors: the repo root, and the writer's own directory when the author spelled a ./ or ../ prefix. A gate that lives inside a package binds a third at module top -- join(dirname(fileURLToPath(import.meta.url)), '..') -- and every literal it writes against that binding was read from the REPO root, where nothing of the sort exists. check:generated's five artifact-ledger literals are the confirmed instance: all five alive under packages/spec/ and all five printed by the residue as dead leads, under a reason text asserting they were never repo paths. Measured first, as the card required. Over the 201 discovered families' 195 scannable gate scripts, 8 bind a package root this way, 2 of them carry a hint dead at the repo root, 6 such hints exist, 5 re-anchor onto something the tree has and 1 is refused because it reaches nothing. Price: 142089 -> 142139 watch-hint (gate, file) pairs (+50, 0 lost, 0 re-attributed), one family's hint set moves, and 5 of 5 admitted hints are true leads verified at their declaration site. The admitted class is exactly the one whose residue sentence was false: a hint is re-anchored only when no tracked path begins with even its first segment. The layout-moved class is left to triage, admission is untouched, and a re-anchoring that reaches nothing is dropped, so the rule cannot invent a path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

unreachableReason's by-construction branch ended "never was a repo path" -- a claim about EVERY base, made from evidence about ONE. check:generated's api-surface, export-origins and src/meta-spelling/... all printed under it while sitting on disk under packages/spec/: a dead-lead label on a live path, telling the reader there is nothing to chase about three files they could open. The sentence now names what the sweep actually established (no tracked path under the literal's first segment) and the two ways that happens, without asserting the literal never existed anywhere. The three docblocks that stated the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep for it still lands on the incident. The pins move with the text rather than staying on the retired phrase: pinned to a string nothing renders, the dist-freshness case would pass on any tree at all -- a green over the exact regression it exists to catch -- so it now asserts against the sentence that branch prints today, and the sweep case asserts both halves. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM ... and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the LAST key in coveringKey, which is an identity claim about a single FILE and never a population. Re-derived at this PR's base: the read edge is 12 families, 16 (family,target) pairs over 15 distinct targets, and it carries 0 inherited hints -- against 373 for the import edge, 7 for the manifest edge and 1 for the run edge. Reading a program's TEXT is not performing its work, so identity-only is the correct behaviour and the sentence was the defect. No behaviour change. The two numberings inside the file also disagreed: discoverFamilies called the spawn follow the SECOND edge and the manifest follow the THIRD, while the docblock called the manifest one the FOURTH. They count different things -- source-order edges, and population follows -- so each now says which, and the claim that there are three population follows all ending at a program is gone. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

…spatch-gates-package-root-anchor

claude · 2026-09-02T07:27:23Z

Fixes#14208
Fixes#14289

Two deliverables in one PR on one file, scripts/pm/dispatch-gates.mjs, as three commits: the package-root anchor (#14208), the residue reason text it is coupled to (#14208), and the read-edge prose repair (#14289, no behaviour change).

#14208 — measure first, as the card required

Premise re-confirmed on this PR's base (ad54eb342), with the card's own command:

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --residue packages/spec/src/index.ts

check:generated printed 5 of 10 declared literal(s) reach nothing tracked, the three named literals among them, each under the reason no tracked path under its first segment; never was a repo path — while all three sit on disk under packages/spec/.

The population sweep

Over the 201 discovered families' 195 scannable gate scripts, resolving every import.meta-seeded binding through the tool's own resolvePathExpression:

measurement	value
discovered gate scripts binding a package root	8 (all `packages/spec`)
...of those, scripts carrying a hint dead at the repo root	2
dead-at-root hints in that population	6
...that re-anchor onto something the tree HAS	5
...refused because the re-anchored form reaches nothing	1 (`'zod'`, a package specifier)

⚠️The card's description of the shape does not survive the sweep, and this changed the rule. The card described the class as literals passed to resolve(PKG_DIR, …). That shape is real — 6 such literals across 6 of the 8 scripts, every one a tracked file — but it is not where the confirmed instance comes from. All five of check:generated's dead literals are artifact: values in its GATED ledger (artifact: 'api-surface/', artifact: 'src/meta-spelling/meta-url-data.generated.ts', …), describing what each gate verifies; nothing resolves them. A rule restricted to the syntactic shape would have repaired none of the three dead leads the card names. The restriction was measured and dropped; the reachability test is what keeps the wider reading honest.

The rule that landed

A hint is re-read against the package root its writer binds only when deepestTrackedPrefix is empty — no tracked path begins with even its first segment, which is exactly the branch whose reason text was false — and only if the re-anchored form reaches the tree through hintReachesTree. It replaces rather than adds. Neither priced refusal is relaxed and neither is reachable from here: admission is untouched (a literal that is not already a hint never arrives), and a re-anchoring that names nothing is dropped, so the rule cannot invent a path. The layout-moved class (a shorter surviving prefix) is deliberately left alone — its first segment is a real repo directory, so re-anchoring it would replace a triage lead with a fabricated one.

The price, before and after, on this base

Over 201 discovered families and 7900 tracked files:

	before	after
watch-hint (gate, file) pairs	142089	142139 (+50, 0 lost)
families whose hint set moves	—	1 (`check:generated`, no other)
pairs re-attributed (same pair, new via)	—	0 — every replaced spelling was dead, so it covered nothing to re-attribute
(check, hint) live / inert	1108 / 152	1113 / 147
distinct hints in the fleet	725	722 — three of the five re-anchored forms are already spelled from the root by other gates
families carrying dead literals	46	45
dead literals	152	147
...in the empty-`deepest` class	128	123

Provenance, which is the criterion this file prices and never volume: 5 of 5 admitted hints are TRUE leads, and 50 of 50 pairs with them, each verified at its declaration site — they are check:generated's own artifact ledger and verifying those artifacts is the gate's whole job:

re-anchored hint	files	ledger entry
`packages/spec/api-surface`	17	`check:api-surface`
`packages/spec/export-origins`	17	`check:export-origins`
`packages/spec/declaration-map`	14	`check:declaration-map`
`packages/spec/src/meta-spelling/meta-url-data.generated.ts`	1	`check:meta-url-spelling`
`packages/spec/liveness/state-counts.md`	1	the liveness counts

0 fabricated. For scale, hintCovers' docblock prices the bare-top-level-word admission it refuses at +139084 pairs on this corpus; this is 0.04% of it, and moduleRelativeDirectoryHint's naive-resolved-form widening was +53 hints for 1 true lead against this rule's 5 for 5.

The whole --residue output moves by exactly one row before and after (diffed line by line).

The reason text

The by-construction sentence ended never was a repo path — a claim about EVERY base made from evidence about ONE, and a dead-lead label printed over three live files. It now reads:

'HINT' — no tracked path under its first segment; a specifier, or a path anchored at a base this scan did not resolve

Checkable by the reader (git ls-files on the first segment is empty) and no longer asserting a base it did not test. The three docblocks stating the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep still lands on the incident. The pins move with the text: pinned to a retired phrase, the dist-freshness case would pass on any tree at all — a green over the exact regression it exists to catch.

Ablation

Anchor call replaced with return [...hints] on the committed tree, mutation proven on disk before the run (anchor occurrences 1 to 0, injected marker 1, blob 9743275c to e51f9d24), restore leg trapped and then proven by byte comparison (git diff HEAD empty, blob back to the HEAD blob 9743275c). Ablated, check:generated returns to 5 of 10 declared literal(s) reach nothing tracked with the three named literals marked. The ablated run also shows the corrected reason text in place, so the two halves are independent — the reason-text repair is the deliverable the card specified for the no-repair branch, and it stands on its own.

#14289 — prose only, no behaviour change

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM … and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the last key in coveringKey, an identity claim about a single FILE (#13000).

Re-derived at this PR's base, not copied from the card:

EDGE import: 27 families, 38 (family,target) pairs, 15 distinct targets
EDGE run: 1 family, 1 pair, 1 distinct target
EDGE manifest: 10 families, 10 pairs, 2 distinct targets
EDGE read: 12 families, 16 pairs, 15 distinct targets
INHERITED HINTS BY EDGE LABEL:
import: 373 hints across 13 families
manifest: 7 hints across 7 families
run: 1 hint across 1 family
read: 0

Read targets that carried a hint into their family: 0. Of the 16 read-edge pairs, 8 are answered by the READ identity key and 8 by an earlier key — consistent with coveringKey's "additive by construction" note.

The two numberings are reconciled rather than flattened: the docblock numbers all four scans in source order (so the manifest edge is the fourth), discoverFamilies numbers only the population follows (so the same edge is its third), and each now says which. The claim that there are three population follows all ending at a program is gone, and packageManifestTargets' "the other three follows take" is corrected to two.

Gates

Union re-run after the final commit, at git rev-parse --short HEAD = 70c1aedff (merge of origin/maind16df7418). Derived with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands — 15 families, all named by path, 0 by kind; the change set is exactly the one file. Exit codes captured to a file before any pipe.

gate	exit	verdict
`pnpm check:pm-dispatch-gates` (the tool's `--self-test`)	0	RAN-PASS — `dispatch-gates self-test: 1240 cases pass`
`node scripts/check-ci-filter-parity.mjs`	0	RAN-PASS
`node scripts/check-cross-package-test-inputs.mjs` / `pnpm check:cross-package-test-inputs`	0 / 0	RAN-PASS (both spellings)
`node scripts/check-self-test-wired.mjs`	0	RAN-PASS
`node scripts/check-shard-attestation.mjs`	0	RAN-PASS
`node scripts/check-test-completeness.mjs`	3	NOT MEASURED — no test log to hand it; the gate says so itself and calls exit 3 distinct from a finding's 1
`pnpm check:agent-test-spelling`	0	RAN-PASS
`pnpm check:bash32-floor`	0	RAN-PASS
`pnpm check:cli-command-ids`	0	RAN-PASS
`pnpm check:declared-population-live`	0	RAN-PASS
`pnpm check:entry-guard`	0	RAN-PASS
`pnpm check:parse-guard`	0	RAN-PASS
`pnpm check:pnpm-filter-targets`	0	RAN-PASS
`pnpm check:watch-hint-literal`	0	RAN-PASS
`pnpm check:ratchet-remedy-authority`	0	RAN-PASS (beyond the derived family)
`node scripts/check-self-test-workflow-commands.mjs`	0	RAN-PASS (beyond the derived family)
`pnpm check:nul-bytes`	0	RAN-PASS

Lint, narrowed and declared.pnpm lint is eslint . --no-inline-config over the whole repo; the run here is targeted at the one changed file: 1 file linted, 0 errors, 0 warnings, exit 0, the count read from --format json. The narrowing is a measurement rather than a gap because this repo runs one eslint.config.mjs which never enables type-aware linting for any file (no parserOptions.project, no typed rules — stated and positive-control-measured in that config's own header), so a one-file diff cannot move the verdict on any untouched file, and the file is a config match rather than an ignore (it produced a result object, not an ignore warning).

Changeset: none. This is a scripts/**-only diff that publishes nothing from any package, which is the repo convention on recent scripts-only PRs; the skip-changeset label carries it.

Generated by Claude Code

…iter binds extractWatchHints resolves a path literal against exactly two anchors: the repo root, and the writer's own directory when the author spelled a ./ or ../ prefix. A gate that lives inside a package binds a third at module top -- join(dirname(fileURLToPath(import.meta.url)), '..') -- and every literal it writes against that binding was read from the REPO root, where nothing of the sort exists. check:generated's five artifact-ledger literals are the confirmed instance: all five alive under packages/spec/ and all five printed by the residue as dead leads, under a reason text asserting they were never repo paths. Measured first, as the card required. Over the 201 discovered families' 195 scannable gate scripts, 8 bind a package root this way, 2 of them carry a hint dead at the repo root, 6 such hints exist, 5 re-anchor onto something the tree has and 1 is refused because it reaches nothing. Price: 142089 -> 142139 watch-hint (gate, file) pairs (+50, 0 lost, 0 re-attributed), one family's hint set moves, and 5 of 5 admitted hints are true leads verified at their declaration site. The admitted class is exactly the one whose residue sentence was false: a hint is re-anchored only when no tracked path begins with even its first segment. The layout-moved class is left to triage, admission is untouched, and a re-anchoring that reaches nothing is dropped, so the rule cannot invent a path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

unreachableReason's by-construction branch ended "never was a repo path" -- a claim about EVERY base, made from evidence about ONE. check:generated's api-surface, export-origins and src/meta-spelling/... all printed under it while sitting on disk under packages/spec/: a dead-lead label on a live path, telling the reader there is nothing to chase about three files they could open. The sentence now names what the sweep actually established (no tracked path under the literal's first segment) and the two ways that happens, without asserting the literal never existed anywhere. The three docblocks that stated the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep for it still lands on the incident. The pins move with the text rather than staying on the retired phrase: pinned to a string nothing renders, the dist-freshness case would pass on any tree at all -- a green over the exact regression it exists to catch -- so it now asserts against the sentence that branch prints today, and the sweep case asserts both halves. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM ... and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the LAST key in coveringKey, which is an identity claim about a single FILE and never a population. Re-derived at this PR's base: the read edge is 12 families, 16 (family,target) pairs over 15 distinct targets, and it carries 0 inherited hints -- against 373 for the import edge, 7 for the manifest edge and 1 for the run edge. Reading a program's TEXT is not performing its work, so identity-only is the correct behaviour and the sentence was the defect. No behaviour change. The two numberings inside the file also disagreed: discoverFamilies called the spawn follow the SECOND edge and the manifest follow the THIRD, while the docblock called the manifest one the FOURTH. They count different things -- source-order edges, and population follows -- so each now says which, and the claim that there are three population follows all ending at a program is gone. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

…spatch-gates-package-root-anchor

claude · 2026-09-02T07:27:23Z

Fixes#14208
Fixes#14289

Two deliverables in one PR on one file, scripts/pm/dispatch-gates.mjs, as three commits: the package-root anchor (#14208), the residue reason text it is coupled to (#14208), and the read-edge prose repair (#14289, no behaviour change).

#14208 — measure first, as the card required

Premise re-confirmed on this PR's base (ad54eb342), with the card's own command:

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --residue packages/spec/src/index.ts

check:generated printed 5 of 10 declared literal(s) reach nothing tracked, the three named literals among them, each under the reason no tracked path under its first segment; never was a repo path — while all three sit on disk under packages/spec/.

The population sweep

Over the 201 discovered families' 195 scannable gate scripts, resolving every import.meta-seeded binding through the tool's own resolvePathExpression:

measurement	value
discovered gate scripts binding a package root	8 (all `packages/spec`)
...of those, scripts carrying a hint dead at the repo root	2
dead-at-root hints in that population	6
...that re-anchor onto something the tree HAS	5
...refused because the re-anchored form reaches nothing	1 (`'zod'`, a package specifier)

⚠️The card's description of the shape does not survive the sweep, and this changed the rule. The card described the class as literals passed to resolve(PKG_DIR, …). That shape is real — 6 such literals across 6 of the 8 scripts, every one a tracked file — but it is not where the confirmed instance comes from. All five of check:generated's dead literals are artifact: values in its GATED ledger (artifact: 'api-surface/', artifact: 'src/meta-spelling/meta-url-data.generated.ts', …), describing what each gate verifies; nothing resolves them. A rule restricted to the syntactic shape would have repaired none of the three dead leads the card names. The restriction was measured and dropped; the reachability test is what keeps the wider reading honest.

The rule that landed

A hint is re-read against the package root its writer binds only when deepestTrackedPrefix is empty — no tracked path begins with even its first segment, which is exactly the branch whose reason text was false — and only if the re-anchored form reaches the tree through hintReachesTree. It replaces rather than adds. Neither priced refusal is relaxed and neither is reachable from here: admission is untouched (a literal that is not already a hint never arrives), and a re-anchoring that names nothing is dropped, so the rule cannot invent a path. The layout-moved class (a shorter surviving prefix) is deliberately left alone — its first segment is a real repo directory, so re-anchoring it would replace a triage lead with a fabricated one.

The price, before and after, on this base

Over 201 discovered families and 7900 tracked files:

	before	after
watch-hint (gate, file) pairs	142089	142139 (+50, 0 lost)
families whose hint set moves	—	1 (`check:generated`, no other)
pairs re-attributed (same pair, new via)	—	0 — every replaced spelling was dead, so it covered nothing to re-attribute
(check, hint) live / inert	1108 / 152	1113 / 147
distinct hints in the fleet	725	722 — three of the five re-anchored forms are already spelled from the root by other gates
families carrying dead literals	46	45
dead literals	152	147
...in the empty-`deepest` class	128	123

Provenance, which is the criterion this file prices and never volume: 5 of 5 admitted hints are TRUE leads, and 50 of 50 pairs with them, each verified at its declaration site — they are check:generated's own artifact ledger and verifying those artifacts is the gate's whole job:

re-anchored hint	files	ledger entry
`packages/spec/api-surface`	17	`check:api-surface`
`packages/spec/export-origins`	17	`check:export-origins`
`packages/spec/declaration-map`	14	`check:declaration-map`
`packages/spec/src/meta-spelling/meta-url-data.generated.ts`	1	`check:meta-url-spelling`
`packages/spec/liveness/state-counts.md`	1	the liveness counts

0 fabricated. For scale, hintCovers' docblock prices the bare-top-level-word admission it refuses at +139084 pairs on this corpus; this is 0.04% of it, and moduleRelativeDirectoryHint's naive-resolved-form widening was +53 hints for 1 true lead against this rule's 5 for 5.

The whole --residue output moves by exactly one row before and after (diffed line by line).

The reason text

The by-construction sentence ended never was a repo path — a claim about EVERY base made from evidence about ONE, and a dead-lead label printed over three live files. It now reads:

'HINT' — no tracked path under its first segment; a specifier, or a path anchored at a base this scan did not resolve

Checkable by the reader (git ls-files on the first segment is empty) and no longer asserting a base it did not test. The three docblocks stating the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep still lands on the incident. The pins move with the text: pinned to a retired phrase, the dist-freshness case would pass on any tree at all — a green over the exact regression it exists to catch.

Ablation

Anchor call replaced with return [...hints] on the committed tree, mutation proven on disk before the run (anchor occurrences 1 to 0, injected marker 1, blob 9743275c to e51f9d24), restore leg trapped and then proven by byte comparison (git diff HEAD empty, blob back to the HEAD blob 9743275c). Ablated, check:generated returns to 5 of 10 declared literal(s) reach nothing tracked with the three named literals marked. The ablated run also shows the corrected reason text in place, so the two halves are independent — the reason-text repair is the deliverable the card specified for the no-repair branch, and it stands on its own.

#14289 — prose only, no behaviour change

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM … and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the last key in coveringKey, an identity claim about a single FILE (#13000).

Re-derived at this PR's base, not copied from the card:

EDGE import: 27 families, 38 (family,target) pairs, 15 distinct targets
EDGE run: 1 family, 1 pair, 1 distinct target
EDGE manifest: 10 families, 10 pairs, 2 distinct targets
EDGE read: 12 families, 16 pairs, 15 distinct targets
INHERITED HINTS BY EDGE LABEL:
import: 373 hints across 13 families
manifest: 7 hints across 7 families
run: 1 hint across 1 family
read: 0

Read targets that carried a hint into their family: 0. Of the 16 read-edge pairs, 8 are answered by the READ identity key and 8 by an earlier key — consistent with coveringKey's "additive by construction" note.

The two numberings are reconciled rather than flattened: the docblock numbers all four scans in source order (so the manifest edge is the fourth), discoverFamilies numbers only the population follows (so the same edge is its third), and each now says which. The claim that there are three population follows all ending at a program is gone, and packageManifestTargets' "the other three follows take" is corrected to two.

Gates

Union re-run after the final commit, at git rev-parse --short HEAD = 70c1aedff (merge of origin/maind16df7418). Derived with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands — 15 families, all named by path, 0 by kind; the change set is exactly the one file. Exit codes captured to a file before any pipe.

gate	exit	verdict
`pnpm check:pm-dispatch-gates` (the tool's `--self-test`)	0	RAN-PASS — `dispatch-gates self-test: 1240 cases pass`
`node scripts/check-ci-filter-parity.mjs`	0	RAN-PASS
`node scripts/check-cross-package-test-inputs.mjs` / `pnpm check:cross-package-test-inputs`	0 / 0	RAN-PASS (both spellings)
`node scripts/check-self-test-wired.mjs`	0	RAN-PASS
`node scripts/check-shard-attestation.mjs`	0	RAN-PASS
`node scripts/check-test-completeness.mjs`	3	NOT MEASURED — no test log to hand it; the gate says so itself and calls exit 3 distinct from a finding's 1
`pnpm check:agent-test-spelling`	0	RAN-PASS
`pnpm check:bash32-floor`	0	RAN-PASS
`pnpm check:cli-command-ids`	0	RAN-PASS
`pnpm check:declared-population-live`	0	RAN-PASS
`pnpm check:entry-guard`	0	RAN-PASS
`pnpm check:parse-guard`	0	RAN-PASS
`pnpm check:pnpm-filter-targets`	0	RAN-PASS
`pnpm check:watch-hint-literal`	0	RAN-PASS
`pnpm check:ratchet-remedy-authority`	0	RAN-PASS (beyond the derived family)
`node scripts/check-self-test-workflow-commands.mjs`	0	RAN-PASS (beyond the derived family)
`pnpm check:nul-bytes`	0	RAN-PASS

Lint, narrowed and declared.pnpm lint is eslint . --no-inline-config over the whole repo; the run here is targeted at the one changed file: 1 file linted, 0 errors, 0 warnings, exit 0, the count read from --format json. The narrowing is a measurement rather than a gap because this repo runs one eslint.config.mjs which never enables type-aware linting for any file (no parserOptions.project, no typed rules — stated and positive-control-measured in that config's own header), so a one-file diff cannot move the verdict on any untouched file, and the file is a config match rather than an ignore (it produced a result object, not an ignore warning).

Changeset: none. This is a scripts/**-only diff that publishes nothing from any package, which is the repo convention on recent scripts-only PRs; the skip-changeset label carries it.

Generated by Claude Code

…iter binds extractWatchHints resolves a path literal against exactly two anchors: the repo root, and the writer's own directory when the author spelled a ./ or ../ prefix. A gate that lives inside a package binds a third at module top -- join(dirname(fileURLToPath(import.meta.url)), '..') -- and every literal it writes against that binding was read from the REPO root, where nothing of the sort exists. check:generated's five artifact-ledger literals are the confirmed instance: all five alive under packages/spec/ and all five printed by the residue as dead leads, under a reason text asserting they were never repo paths. Measured first, as the card required. Over the 201 discovered families' 195 scannable gate scripts, 8 bind a package root this way, 2 of them carry a hint dead at the repo root, 6 such hints exist, 5 re-anchor onto something the tree has and 1 is refused because it reaches nothing. Price: 142089 -> 142139 watch-hint (gate, file) pairs (+50, 0 lost, 0 re-attributed), one family's hint set moves, and 5 of 5 admitted hints are true leads verified at their declaration site. The admitted class is exactly the one whose residue sentence was false: a hint is re-anchored only when no tracked path begins with even its first segment. The layout-moved class is left to triage, admission is untouched, and a re-anchoring that reaches nothing is dropped, so the rule cannot invent a path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

unreachableReason's by-construction branch ended "never was a repo path" -- a claim about EVERY base, made from evidence about ONE. check:generated's api-surface, export-origins and src/meta-spelling/... all printed under it while sitting on disk under packages/spec/: a dead-lead label on a live path, telling the reader there is nothing to chase about three files they could open. The sentence now names what the sweep actually established (no tracked path under the literal's first segment) and the two ways that happens, without asserting the literal never existed anywhere. The three docblocks that stated the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep for it still lands on the incident. The pins move with the text rather than staying on the retired phrase: pinned to a string nothing renders, the dist-freshness case would pass on any tree at all -- a green over the exact regression it exists to catch -- so it now asserts against the sentence that branch prints today, and the sweep case asserts both halves. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM ... and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the LAST key in coveringKey, which is an identity claim about a single FILE and never a population. Re-derived at this PR's base: the read edge is 12 families, 16 (family,target) pairs over 15 distinct targets, and it carries 0 inherited hints -- against 373 for the import edge, 7 for the manifest edge and 1 for the run edge. Reading a program's TEXT is not performing its work, so identity-only is the correct behaviour and the sentence was the defect. No behaviour change. The two numberings inside the file also disagreed: discoverFamilies called the spawn follow the SECOND edge and the manifest follow the THIRD, while the docblock called the manifest one the FOURTH. They count different things -- source-order edges, and population follows -- so each now says which, and the claim that there are three population follows all ending at a program is gone. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

…spatch-gates-package-root-anchor

claude · 2026-09-02T07:27:23Z

Fixes#14208
Fixes#14289

Two deliverables in one PR on one file, scripts/pm/dispatch-gates.mjs, as three commits: the package-root anchor (#14208), the residue reason text it is coupled to (#14208), and the read-edge prose repair (#14289, no behaviour change).

#14208 — measure first, as the card required

Premise re-confirmed on this PR's base (ad54eb342), with the card's own command:

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --residue packages/spec/src/index.ts

check:generated printed 5 of 10 declared literal(s) reach nothing tracked, the three named literals among them, each under the reason no tracked path under its first segment; never was a repo path — while all three sit on disk under packages/spec/.

The population sweep

Over the 201 discovered families' 195 scannable gate scripts, resolving every import.meta-seeded binding through the tool's own resolvePathExpression:

measurement	value
discovered gate scripts binding a package root	8 (all `packages/spec`)
...of those, scripts carrying a hint dead at the repo root	2
dead-at-root hints in that population	6
...that re-anchor onto something the tree HAS	5
...refused because the re-anchored form reaches nothing	1 (`'zod'`, a package specifier)

⚠️The card's description of the shape does not survive the sweep, and this changed the rule. The card described the class as literals passed to resolve(PKG_DIR, …). That shape is real — 6 such literals across 6 of the 8 scripts, every one a tracked file — but it is not where the confirmed instance comes from. All five of check:generated's dead literals are artifact: values in its GATED ledger (artifact: 'api-surface/', artifact: 'src/meta-spelling/meta-url-data.generated.ts', …), describing what each gate verifies; nothing resolves them. A rule restricted to the syntactic shape would have repaired none of the three dead leads the card names. The restriction was measured and dropped; the reachability test is what keeps the wider reading honest.

The rule that landed

A hint is re-read against the package root its writer binds only when deepestTrackedPrefix is empty — no tracked path begins with even its first segment, which is exactly the branch whose reason text was false — and only if the re-anchored form reaches the tree through hintReachesTree. It replaces rather than adds. Neither priced refusal is relaxed and neither is reachable from here: admission is untouched (a literal that is not already a hint never arrives), and a re-anchoring that names nothing is dropped, so the rule cannot invent a path. The layout-moved class (a shorter surviving prefix) is deliberately left alone — its first segment is a real repo directory, so re-anchoring it would replace a triage lead with a fabricated one.

The price, before and after, on this base

Over 201 discovered families and 7900 tracked files:

	before	after
watch-hint (gate, file) pairs	142089	142139 (+50, 0 lost)
families whose hint set moves	—	1 (`check:generated`, no other)
pairs re-attributed (same pair, new via)	—	0 — every replaced spelling was dead, so it covered nothing to re-attribute
(check, hint) live / inert	1108 / 152	1113 / 147
distinct hints in the fleet	725	722 — three of the five re-anchored forms are already spelled from the root by other gates
families carrying dead literals	46	45
dead literals	152	147
...in the empty-`deepest` class	128	123

Provenance, which is the criterion this file prices and never volume: 5 of 5 admitted hints are TRUE leads, and 50 of 50 pairs with them, each verified at its declaration site — they are check:generated's own artifact ledger and verifying those artifacts is the gate's whole job:

re-anchored hint	files	ledger entry
`packages/spec/api-surface`	17	`check:api-surface`
`packages/spec/export-origins`	17	`check:export-origins`
`packages/spec/declaration-map`	14	`check:declaration-map`
`packages/spec/src/meta-spelling/meta-url-data.generated.ts`	1	`check:meta-url-spelling`
`packages/spec/liveness/state-counts.md`	1	the liveness counts

0 fabricated. For scale, hintCovers' docblock prices the bare-top-level-word admission it refuses at +139084 pairs on this corpus; this is 0.04% of it, and moduleRelativeDirectoryHint's naive-resolved-form widening was +53 hints for 1 true lead against this rule's 5 for 5.

The whole --residue output moves by exactly one row before and after (diffed line by line).

The reason text

The by-construction sentence ended never was a repo path — a claim about EVERY base made from evidence about ONE, and a dead-lead label printed over three live files. It now reads:

'HINT' — no tracked path under its first segment; a specifier, or a path anchored at a base this scan did not resolve

Checkable by the reader (git ls-files on the first segment is empty) and no longer asserting a base it did not test. The three docblocks stating the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep still lands on the incident. The pins move with the text: pinned to a retired phrase, the dist-freshness case would pass on any tree at all — a green over the exact regression it exists to catch.

Ablation

Anchor call replaced with return [...hints] on the committed tree, mutation proven on disk before the run (anchor occurrences 1 to 0, injected marker 1, blob 9743275c to e51f9d24), restore leg trapped and then proven by byte comparison (git diff HEAD empty, blob back to the HEAD blob 9743275c). Ablated, check:generated returns to 5 of 10 declared literal(s) reach nothing tracked with the three named literals marked. The ablated run also shows the corrected reason text in place, so the two halves are independent — the reason-text repair is the deliverable the card specified for the no-repair branch, and it stands on its own.

#14289 — prose only, no behaviour change

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM … and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the last key in coveringKey, an identity claim about a single FILE (#13000).

Re-derived at this PR's base, not copied from the card:

EDGE import: 27 families, 38 (family,target) pairs, 15 distinct targets
EDGE run: 1 family, 1 pair, 1 distinct target
EDGE manifest: 10 families, 10 pairs, 2 distinct targets
EDGE read: 12 families, 16 pairs, 15 distinct targets
INHERITED HINTS BY EDGE LABEL:
import: 373 hints across 13 families
manifest: 7 hints across 7 families
run: 1 hint across 1 family
read: 0

Read targets that carried a hint into their family: 0. Of the 16 read-edge pairs, 8 are answered by the READ identity key and 8 by an earlier key — consistent with coveringKey's "additive by construction" note.

The two numberings are reconciled rather than flattened: the docblock numbers all four scans in source order (so the manifest edge is the fourth), discoverFamilies numbers only the population follows (so the same edge is its third), and each now says which. The claim that there are three population follows all ending at a program is gone, and packageManifestTargets' "the other three follows take" is corrected to two.

Gates

Union re-run after the final commit, at git rev-parse --short HEAD = 70c1aedff (merge of origin/maind16df7418). Derived with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands — 15 families, all named by path, 0 by kind; the change set is exactly the one file. Exit codes captured to a file before any pipe.

gate	exit	verdict
`pnpm check:pm-dispatch-gates` (the tool's `--self-test`)	0	RAN-PASS — `dispatch-gates self-test: 1240 cases pass`
`node scripts/check-ci-filter-parity.mjs`	0	RAN-PASS
`node scripts/check-cross-package-test-inputs.mjs` / `pnpm check:cross-package-test-inputs`	0 / 0	RAN-PASS (both spellings)
`node scripts/check-self-test-wired.mjs`	0	RAN-PASS
`node scripts/check-shard-attestation.mjs`	0	RAN-PASS
`node scripts/check-test-completeness.mjs`	3	NOT MEASURED — no test log to hand it; the gate says so itself and calls exit 3 distinct from a finding's 1
`pnpm check:agent-test-spelling`	0	RAN-PASS
`pnpm check:bash32-floor`	0	RAN-PASS
`pnpm check:cli-command-ids`	0	RAN-PASS
`pnpm check:declared-population-live`	0	RAN-PASS
`pnpm check:entry-guard`	0	RAN-PASS
`pnpm check:parse-guard`	0	RAN-PASS
`pnpm check:pnpm-filter-targets`	0	RAN-PASS
`pnpm check:watch-hint-literal`	0	RAN-PASS
`pnpm check:ratchet-remedy-authority`	0	RAN-PASS (beyond the derived family)
`node scripts/check-self-test-workflow-commands.mjs`	0	RAN-PASS (beyond the derived family)
`pnpm check:nul-bytes`	0	RAN-PASS

Lint, narrowed and declared.pnpm lint is eslint . --no-inline-config over the whole repo; the run here is targeted at the one changed file: 1 file linted, 0 errors, 0 warnings, exit 0, the count read from --format json. The narrowing is a measurement rather than a gap because this repo runs one eslint.config.mjs which never enables type-aware linting for any file (no parserOptions.project, no typed rules — stated and positive-control-measured in that config's own header), so a one-file diff cannot move the verdict on any untouched file, and the file is a config match rather than an ignore (it produced a result object, not an ignore warning).

Changeset: none. This is a scripts/**-only diff that publishes nothing from any package, which is the repo convention on recent scripts-only PRs; the skip-changeset label carries it.

Generated by Claude Code

…iter binds extractWatchHints resolves a path literal against exactly two anchors: the repo root, and the writer's own directory when the author spelled a ./ or ../ prefix. A gate that lives inside a package binds a third at module top -- join(dirname(fileURLToPath(import.meta.url)), '..') -- and every literal it writes against that binding was read from the REPO root, where nothing of the sort exists. check:generated's five artifact-ledger literals are the confirmed instance: all five alive under packages/spec/ and all five printed by the residue as dead leads, under a reason text asserting they were never repo paths. Measured first, as the card required. Over the 201 discovered families' 195 scannable gate scripts, 8 bind a package root this way, 2 of them carry a hint dead at the repo root, 6 such hints exist, 5 re-anchor onto something the tree has and 1 is refused because it reaches nothing. Price: 142089 -> 142139 watch-hint (gate, file) pairs (+50, 0 lost, 0 re-attributed), one family's hint set moves, and 5 of 5 admitted hints are true leads verified at their declaration site. The admitted class is exactly the one whose residue sentence was false: a hint is re-anchored only when no tracked path begins with even its first segment. The layout-moved class is left to triage, admission is untouched, and a re-anchoring that reaches nothing is dropped, so the rule cannot invent a path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

unreachableReason's by-construction branch ended "never was a repo path" -- a claim about EVERY base, made from evidence about ONE. check:generated's api-surface, export-origins and src/meta-spelling/... all printed under it while sitting on disk under packages/spec/: a dead-lead label on a live path, telling the reader there is nothing to chase about three files they could open. The sentence now names what the sweep actually established (no tracked path under the literal's first segment) and the two ways that happens, without asserting the literal never existed anywhere. The three docblocks that stated the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep for it still lands on the incident. The pins move with the text rather than staying on the retired phrase: pinned to a string nothing renders, the dist-freshness case would pass on any tree at all -- a green over the exact regression it exists to catch -- so it now asserts against the sentence that branch prints today, and the sweep case asserts both halves. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM ... and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the LAST key in coveringKey, which is an identity claim about a single FILE and never a population. Re-derived at this PR's base: the read edge is 12 families, 16 (family,target) pairs over 15 distinct targets, and it carries 0 inherited hints -- against 373 for the import edge, 7 for the manifest edge and 1 for the run edge. Reading a program's TEXT is not performing its work, so identity-only is the correct behaviour and the sentence was the defect. No behaviour change. The two numberings inside the file also disagreed: discoverFamilies called the spawn follow the SECOND edge and the manifest follow the THIRD, while the docblock called the manifest one the FOURTH. They count different things -- source-order edges, and population follows -- so each now says which, and the claim that there are three population follows all ending at a program is gone. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

…spatch-gates-package-root-anchor

claude · 2026-09-02T07:27:23Z

Fixes#14208
Fixes#14289

Two deliverables in one PR on one file, scripts/pm/dispatch-gates.mjs, as three commits: the package-root anchor (#14208), the residue reason text it is coupled to (#14208), and the read-edge prose repair (#14289, no behaviour change).

#14208 — measure first, as the card required

Premise re-confirmed on this PR's base (ad54eb342), with the card's own command:

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --residue packages/spec/src/index.ts

check:generated printed 5 of 10 declared literal(s) reach nothing tracked, the three named literals among them, each under the reason no tracked path under its first segment; never was a repo path — while all three sit on disk under packages/spec/.

The population sweep

Over the 201 discovered families' 195 scannable gate scripts, resolving every import.meta-seeded binding through the tool's own resolvePathExpression:

measurement	value
discovered gate scripts binding a package root	8 (all `packages/spec`)
...of those, scripts carrying a hint dead at the repo root	2
dead-at-root hints in that population	6
...that re-anchor onto something the tree HAS	5
...refused because the re-anchored form reaches nothing	1 (`'zod'`, a package specifier)

⚠️The card's description of the shape does not survive the sweep, and this changed the rule. The card described the class as literals passed to resolve(PKG_DIR, …). That shape is real — 6 such literals across 6 of the 8 scripts, every one a tracked file — but it is not where the confirmed instance comes from. All five of check:generated's dead literals are artifact: values in its GATED ledger (artifact: 'api-surface/', artifact: 'src/meta-spelling/meta-url-data.generated.ts', …), describing what each gate verifies; nothing resolves them. A rule restricted to the syntactic shape would have repaired none of the three dead leads the card names. The restriction was measured and dropped; the reachability test is what keeps the wider reading honest.

The rule that landed

A hint is re-read against the package root its writer binds only when deepestTrackedPrefix is empty — no tracked path begins with even its first segment, which is exactly the branch whose reason text was false — and only if the re-anchored form reaches the tree through hintReachesTree. It replaces rather than adds. Neither priced refusal is relaxed and neither is reachable from here: admission is untouched (a literal that is not already a hint never arrives), and a re-anchoring that names nothing is dropped, so the rule cannot invent a path. The layout-moved class (a shorter surviving prefix) is deliberately left alone — its first segment is a real repo directory, so re-anchoring it would replace a triage lead with a fabricated one.

The price, before and after, on this base

Over 201 discovered families and 7900 tracked files:

	before	after
watch-hint (gate, file) pairs	142089	142139 (+50, 0 lost)
families whose hint set moves	—	1 (`check:generated`, no other)
pairs re-attributed (same pair, new via)	—	0 — every replaced spelling was dead, so it covered nothing to re-attribute
(check, hint) live / inert	1108 / 152	1113 / 147
distinct hints in the fleet	725	722 — three of the five re-anchored forms are already spelled from the root by other gates
families carrying dead literals	46	45
dead literals	152	147
...in the empty-`deepest` class	128	123

Provenance, which is the criterion this file prices and never volume: 5 of 5 admitted hints are TRUE leads, and 50 of 50 pairs with them, each verified at its declaration site — they are check:generated's own artifact ledger and verifying those artifacts is the gate's whole job:

re-anchored hint	files	ledger entry
`packages/spec/api-surface`	17	`check:api-surface`
`packages/spec/export-origins`	17	`check:export-origins`
`packages/spec/declaration-map`	14	`check:declaration-map`
`packages/spec/src/meta-spelling/meta-url-data.generated.ts`	1	`check:meta-url-spelling`
`packages/spec/liveness/state-counts.md`	1	the liveness counts

0 fabricated. For scale, hintCovers' docblock prices the bare-top-level-word admission it refuses at +139084 pairs on this corpus; this is 0.04% of it, and moduleRelativeDirectoryHint's naive-resolved-form widening was +53 hints for 1 true lead against this rule's 5 for 5.

The whole --residue output moves by exactly one row before and after (diffed line by line).

The reason text

The by-construction sentence ended never was a repo path — a claim about EVERY base made from evidence about ONE, and a dead-lead label printed over three live files. It now reads:

'HINT' — no tracked path under its first segment; a specifier, or a path anchored at a base this scan did not resolve

Checkable by the reader (git ls-files on the first segment is empty) and no longer asserting a base it did not test. The three docblocks stating the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep still lands on the incident. The pins move with the text: pinned to a retired phrase, the dist-freshness case would pass on any tree at all — a green over the exact regression it exists to catch.

Ablation

Anchor call replaced with return [...hints] on the committed tree, mutation proven on disk before the run (anchor occurrences 1 to 0, injected marker 1, blob 9743275c to e51f9d24), restore leg trapped and then proven by byte comparison (git diff HEAD empty, blob back to the HEAD blob 9743275c). Ablated, check:generated returns to 5 of 10 declared literal(s) reach nothing tracked with the three named literals marked. The ablated run also shows the corrected reason text in place, so the two halves are independent — the reason-text repair is the deliverable the card specified for the no-repair branch, and it stands on its own.

#14289 — prose only, no behaviour change

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM … and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the last key in coveringKey, an identity claim about a single FILE (#13000).

Re-derived at this PR's base, not copied from the card:

EDGE import: 27 families, 38 (family,target) pairs, 15 distinct targets
EDGE run: 1 family, 1 pair, 1 distinct target
EDGE manifest: 10 families, 10 pairs, 2 distinct targets
EDGE read: 12 families, 16 pairs, 15 distinct targets
INHERITED HINTS BY EDGE LABEL:
import: 373 hints across 13 families
manifest: 7 hints across 7 families
run: 1 hint across 1 family
read: 0

Read targets that carried a hint into their family: 0. Of the 16 read-edge pairs, 8 are answered by the READ identity key and 8 by an earlier key — consistent with coveringKey's "additive by construction" note.

The two numberings are reconciled rather than flattened: the docblock numbers all four scans in source order (so the manifest edge is the fourth), discoverFamilies numbers only the population follows (so the same edge is its third), and each now says which. The claim that there are three population follows all ending at a program is gone, and packageManifestTargets' "the other three follows take" is corrected to two.

Gates

Union re-run after the final commit, at git rev-parse --short HEAD = 70c1aedff (merge of origin/maind16df7418). Derived with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands — 15 families, all named by path, 0 by kind; the change set is exactly the one file. Exit codes captured to a file before any pipe.

gate	exit	verdict
`pnpm check:pm-dispatch-gates` (the tool's `--self-test`)	0	RAN-PASS — `dispatch-gates self-test: 1240 cases pass`
`node scripts/check-ci-filter-parity.mjs`	0	RAN-PASS
`node scripts/check-cross-package-test-inputs.mjs` / `pnpm check:cross-package-test-inputs`	0 / 0	RAN-PASS (both spellings)
`node scripts/check-self-test-wired.mjs`	0	RAN-PASS
`node scripts/check-shard-attestation.mjs`	0	RAN-PASS
`node scripts/check-test-completeness.mjs`	3	NOT MEASURED — no test log to hand it; the gate says so itself and calls exit 3 distinct from a finding's 1
`pnpm check:agent-test-spelling`	0	RAN-PASS
`pnpm check:bash32-floor`	0	RAN-PASS
`pnpm check:cli-command-ids`	0	RAN-PASS
`pnpm check:declared-population-live`	0	RAN-PASS
`pnpm check:entry-guard`	0	RAN-PASS
`pnpm check:parse-guard`	0	RAN-PASS
`pnpm check:pnpm-filter-targets`	0	RAN-PASS
`pnpm check:watch-hint-literal`	0	RAN-PASS
`pnpm check:ratchet-remedy-authority`	0	RAN-PASS (beyond the derived family)
`node scripts/check-self-test-workflow-commands.mjs`	0	RAN-PASS (beyond the derived family)
`pnpm check:nul-bytes`	0	RAN-PASS

Lint, narrowed and declared.pnpm lint is eslint . --no-inline-config over the whole repo; the run here is targeted at the one changed file: 1 file linted, 0 errors, 0 warnings, exit 0, the count read from --format json. The narrowing is a measurement rather than a gap because this repo runs one eslint.config.mjs which never enables type-aware linting for any file (no parserOptions.project, no typed rules — stated and positive-control-measured in that config's own header), so a one-file diff cannot move the verdict on any untouched file, and the file is a config match rather than an ignore (it produced a result object, not an ignore warning).

Changeset: none. This is a scripts/**-only diff that publishes nothing from any package, which is the repo convention on recent scripts-only PRs; the skip-changeset label carries it.

Generated by Claude Code

…iter binds extractWatchHints resolves a path literal against exactly two anchors: the repo root, and the writer's own directory when the author spelled a ./ or ../ prefix. A gate that lives inside a package binds a third at module top -- join(dirname(fileURLToPath(import.meta.url)), '..') -- and every literal it writes against that binding was read from the REPO root, where nothing of the sort exists. check:generated's five artifact-ledger literals are the confirmed instance: all five alive under packages/spec/ and all five printed by the residue as dead leads, under a reason text asserting they were never repo paths. Measured first, as the card required. Over the 201 discovered families' 195 scannable gate scripts, 8 bind a package root this way, 2 of them carry a hint dead at the repo root, 6 such hints exist, 5 re-anchor onto something the tree has and 1 is refused because it reaches nothing. Price: 142089 -> 142139 watch-hint (gate, file) pairs (+50, 0 lost, 0 re-attributed), one family's hint set moves, and 5 of 5 admitted hints are true leads verified at their declaration site. The admitted class is exactly the one whose residue sentence was false: a hint is re-anchored only when no tracked path begins with even its first segment. The layout-moved class is left to triage, admission is untouched, and a re-anchoring that reaches nothing is dropped, so the rule cannot invent a path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

unreachableReason's by-construction branch ended "never was a repo path" -- a claim about EVERY base, made from evidence about ONE. check:generated's api-surface, export-origins and src/meta-spelling/... all printed under it while sitting on disk under packages/spec/: a dead-lead label on a live path, telling the reader there is nothing to chase about three files they could open. The sentence now names what the sweep actually established (no tracked path under the literal's first segment) and the two ways that happens, without asserting the literal never existed anywhere. The three docblocks that stated the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep for it still lands on the incident. The pins move with the text rather than staying on the retired phrase: pinned to a string nothing renders, the dist-freshness case would pass on any tree at all -- a green over the exact regression it exists to catch -- so it now asserts against the sentence that branch prints today, and the sweep case asserts both halves. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM ... and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the LAST key in coveringKey, which is an identity claim about a single FILE and never a population. Re-derived at this PR's base: the read edge is 12 families, 16 (family,target) pairs over 15 distinct targets, and it carries 0 inherited hints -- against 373 for the import edge, 7 for the manifest edge and 1 for the run edge. Reading a program's TEXT is not performing its work, so identity-only is the correct behaviour and the sentence was the defect. No behaviour change. The two numberings inside the file also disagreed: discoverFamilies called the spawn follow the SECOND edge and the manifest follow the THIRD, while the docblock called the manifest one the FOURTH. They count different things -- source-order edges, and population follows -- so each now says which, and the claim that there are three population follows all ending at a program is gone. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

…spatch-gates-package-root-anchor

claude · 2026-09-02T07:27:23Z

Fixes#14208
Fixes#14289

Two deliverables in one PR on one file, scripts/pm/dispatch-gates.mjs, as three commits: the package-root anchor (#14208), the residue reason text it is coupled to (#14208), and the read-edge prose repair (#14289, no behaviour change).

#14208 — measure first, as the card required

Premise re-confirmed on this PR's base (ad54eb342), with the card's own command:

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --residue packages/spec/src/index.ts

check:generated printed 5 of 10 declared literal(s) reach nothing tracked, the three named literals among them, each under the reason no tracked path under its first segment; never was a repo path — while all three sit on disk under packages/spec/.

The population sweep

Over the 201 discovered families' 195 scannable gate scripts, resolving every import.meta-seeded binding through the tool's own resolvePathExpression:

measurement	value
discovered gate scripts binding a package root	8 (all `packages/spec`)
...of those, scripts carrying a hint dead at the repo root	2
dead-at-root hints in that population	6
...that re-anchor onto something the tree HAS	5
...refused because the re-anchored form reaches nothing	1 (`'zod'`, a package specifier)

⚠️The card's description of the shape does not survive the sweep, and this changed the rule. The card described the class as literals passed to resolve(PKG_DIR, …). That shape is real — 6 such literals across 6 of the 8 scripts, every one a tracked file — but it is not where the confirmed instance comes from. All five of check:generated's dead literals are artifact: values in its GATED ledger (artifact: 'api-surface/', artifact: 'src/meta-spelling/meta-url-data.generated.ts', …), describing what each gate verifies; nothing resolves them. A rule restricted to the syntactic shape would have repaired none of the three dead leads the card names. The restriction was measured and dropped; the reachability test is what keeps the wider reading honest.

The rule that landed

A hint is re-read against the package root its writer binds only when deepestTrackedPrefix is empty — no tracked path begins with even its first segment, which is exactly the branch whose reason text was false — and only if the re-anchored form reaches the tree through hintReachesTree. It replaces rather than adds. Neither priced refusal is relaxed and neither is reachable from here: admission is untouched (a literal that is not already a hint never arrives), and a re-anchoring that names nothing is dropped, so the rule cannot invent a path. The layout-moved class (a shorter surviving prefix) is deliberately left alone — its first segment is a real repo directory, so re-anchoring it would replace a triage lead with a fabricated one.

The price, before and after, on this base

Over 201 discovered families and 7900 tracked files:

	before	after
watch-hint (gate, file) pairs	142089	142139 (+50, 0 lost)
families whose hint set moves	—	1 (`check:generated`, no other)
pairs re-attributed (same pair, new via)	—	0 — every replaced spelling was dead, so it covered nothing to re-attribute
(check, hint) live / inert	1108 / 152	1113 / 147
distinct hints in the fleet	725	722 — three of the five re-anchored forms are already spelled from the root by other gates
families carrying dead literals	46	45
dead literals	152	147
...in the empty-`deepest` class	128	123

Provenance, which is the criterion this file prices and never volume: 5 of 5 admitted hints are TRUE leads, and 50 of 50 pairs with them, each verified at its declaration site — they are check:generated's own artifact ledger and verifying those artifacts is the gate's whole job:

re-anchored hint	files	ledger entry
`packages/spec/api-surface`	17	`check:api-surface`
`packages/spec/export-origins`	17	`check:export-origins`
`packages/spec/declaration-map`	14	`check:declaration-map`
`packages/spec/src/meta-spelling/meta-url-data.generated.ts`	1	`check:meta-url-spelling`
`packages/spec/liveness/state-counts.md`	1	the liveness counts

0 fabricated. For scale, hintCovers' docblock prices the bare-top-level-word admission it refuses at +139084 pairs on this corpus; this is 0.04% of it, and moduleRelativeDirectoryHint's naive-resolved-form widening was +53 hints for 1 true lead against this rule's 5 for 5.

The whole --residue output moves by exactly one row before and after (diffed line by line).

The reason text

The by-construction sentence ended never was a repo path — a claim about EVERY base made from evidence about ONE, and a dead-lead label printed over three live files. It now reads:

'HINT' — no tracked path under its first segment; a specifier, or a path anchored at a base this scan did not resolve

Checkable by the reader (git ls-files on the first segment is empty) and no longer asserting a base it did not test. The three docblocks stating the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep still lands on the incident. The pins move with the text: pinned to a retired phrase, the dist-freshness case would pass on any tree at all — a green over the exact regression it exists to catch.

Ablation

Anchor call replaced with return [...hints] on the committed tree, mutation proven on disk before the run (anchor occurrences 1 to 0, injected marker 1, blob 9743275c to e51f9d24), restore leg trapped and then proven by byte comparison (git diff HEAD empty, blob back to the HEAD blob 9743275c). Ablated, check:generated returns to 5 of 10 declared literal(s) reach nothing tracked with the three named literals marked. The ablated run also shows the corrected reason text in place, so the two halves are independent — the reason-text repair is the deliverable the card specified for the no-repair branch, and it stands on its own.

#14289 — prose only, no behaviour change

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM … and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the last key in coveringKey, an identity claim about a single FILE (#13000).

Re-derived at this PR's base, not copied from the card:

EDGE import: 27 families, 38 (family,target) pairs, 15 distinct targets
EDGE run: 1 family, 1 pair, 1 distinct target
EDGE manifest: 10 families, 10 pairs, 2 distinct targets
EDGE read: 12 families, 16 pairs, 15 distinct targets
INHERITED HINTS BY EDGE LABEL:
import: 373 hints across 13 families
manifest: 7 hints across 7 families
run: 1 hint across 1 family
read: 0

Read targets that carried a hint into their family: 0. Of the 16 read-edge pairs, 8 are answered by the READ identity key and 8 by an earlier key — consistent with coveringKey's "additive by construction" note.

The two numberings are reconciled rather than flattened: the docblock numbers all four scans in source order (so the manifest edge is the fourth), discoverFamilies numbers only the population follows (so the same edge is its third), and each now says which. The claim that there are three population follows all ending at a program is gone, and packageManifestTargets' "the other three follows take" is corrected to two.

Gates

Union re-run after the final commit, at git rev-parse --short HEAD = 70c1aedff (merge of origin/maind16df7418). Derived with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands — 15 families, all named by path, 0 by kind; the change set is exactly the one file. Exit codes captured to a file before any pipe.

gate	exit	verdict
`pnpm check:pm-dispatch-gates` (the tool's `--self-test`)	0	RAN-PASS — `dispatch-gates self-test: 1240 cases pass`
`node scripts/check-ci-filter-parity.mjs`	0	RAN-PASS
`node scripts/check-cross-package-test-inputs.mjs` / `pnpm check:cross-package-test-inputs`	0 / 0	RAN-PASS (both spellings)
`node scripts/check-self-test-wired.mjs`	0	RAN-PASS
`node scripts/check-shard-attestation.mjs`	0	RAN-PASS
`node scripts/check-test-completeness.mjs`	3	NOT MEASURED — no test log to hand it; the gate says so itself and calls exit 3 distinct from a finding's 1
`pnpm check:agent-test-spelling`	0	RAN-PASS
`pnpm check:bash32-floor`	0	RAN-PASS
`pnpm check:cli-command-ids`	0	RAN-PASS
`pnpm check:declared-population-live`	0	RAN-PASS
`pnpm check:entry-guard`	0	RAN-PASS
`pnpm check:parse-guard`	0	RAN-PASS
`pnpm check:pnpm-filter-targets`	0	RAN-PASS
`pnpm check:watch-hint-literal`	0	RAN-PASS
`pnpm check:ratchet-remedy-authority`	0	RAN-PASS (beyond the derived family)
`node scripts/check-self-test-workflow-commands.mjs`	0	RAN-PASS (beyond the derived family)
`pnpm check:nul-bytes`	0	RAN-PASS

Lint, narrowed and declared.pnpm lint is eslint . --no-inline-config over the whole repo; the run here is targeted at the one changed file: 1 file linted, 0 errors, 0 warnings, exit 0, the count read from --format json. The narrowing is a measurement rather than a gap because this repo runs one eslint.config.mjs which never enables type-aware linting for any file (no parserOptions.project, no typed rules — stated and positive-control-measured in that config's own header), so a one-file diff cannot move the verdict on any untouched file, and the file is a config match rather than an ignore (it produced a result object, not an ignore warning).

Changeset: none. This is a scripts/**-only diff that publishes nothing from any package, which is the repo convention on recent scripts-only PRs; the skip-changeset label carries it.

Generated by Claude Code

…iter binds extractWatchHints resolves a path literal against exactly two anchors: the repo root, and the writer's own directory when the author spelled a ./ or ../ prefix. A gate that lives inside a package binds a third at module top -- join(dirname(fileURLToPath(import.meta.url)), '..') -- and every literal it writes against that binding was read from the REPO root, where nothing of the sort exists. check:generated's five artifact-ledger literals are the confirmed instance: all five alive under packages/spec/ and all five printed by the residue as dead leads, under a reason text asserting they were never repo paths. Measured first, as the card required. Over the 201 discovered families' 195 scannable gate scripts, 8 bind a package root this way, 2 of them carry a hint dead at the repo root, 6 such hints exist, 5 re-anchor onto something the tree has and 1 is refused because it reaches nothing. Price: 142089 -> 142139 watch-hint (gate, file) pairs (+50, 0 lost, 0 re-attributed), one family's hint set moves, and 5 of 5 admitted hints are true leads verified at their declaration site. The admitted class is exactly the one whose residue sentence was false: a hint is re-anchored only when no tracked path begins with even its first segment. The layout-moved class is left to triage, admission is untouched, and a re-anchoring that reaches nothing is dropped, so the rule cannot invent a path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

unreachableReason's by-construction branch ended "never was a repo path" -- a claim about EVERY base, made from evidence about ONE. check:generated's api-surface, export-origins and src/meta-spelling/... all printed under it while sitting on disk under packages/spec/: a dead-lead label on a live path, telling the reader there is nothing to chase about three files they could open. The sentence now names what the sweep actually established (no tracked path under the literal's first segment) and the two ways that happens, without asserting the literal never existed anywhere. The three docblocks that stated the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep for it still lands on the incident. The pins move with the text rather than staying on the retired phrase: pinned to a string nothing renders, the dist-freshness case would pass on any tree at all -- a green over the exact regression it exists to catch -- so it now asserts against the sentence that branch prints today, and the sweep case asserts both halves. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM ... and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the LAST key in coveringKey, which is an identity claim about a single FILE and never a population. Re-derived at this PR's base: the read edge is 12 families, 16 (family,target) pairs over 15 distinct targets, and it carries 0 inherited hints -- against 373 for the import edge, 7 for the manifest edge and 1 for the run edge. Reading a program's TEXT is not performing its work, so identity-only is the correct behaviour and the sentence was the defect. No behaviour change. The two numberings inside the file also disagreed: discoverFamilies called the spawn follow the SECOND edge and the manifest follow the THIRD, while the docblock called the manifest one the FOURTH. They count different things -- source-order edges, and population follows -- so each now says which, and the claim that there are three population follows all ending at a program is gone. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

…spatch-gates-package-root-anchor

claude · 2026-09-02T07:27:23Z

Fixes#14208
Fixes#14289

Two deliverables in one PR on one file, scripts/pm/dispatch-gates.mjs, as three commits: the package-root anchor (#14208), the residue reason text it is coupled to (#14208), and the read-edge prose repair (#14289, no behaviour change).

#14208 — measure first, as the card required

Premise re-confirmed on this PR's base (ad54eb342), with the card's own command:

node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --residue packages/spec/src/index.ts

check:generated printed 5 of 10 declared literal(s) reach nothing tracked, the three named literals among them, each under the reason no tracked path under its first segment; never was a repo path — while all three sit on disk under packages/spec/.

The population sweep

Over the 201 discovered families' 195 scannable gate scripts, resolving every import.meta-seeded binding through the tool's own resolvePathExpression:

measurement	value
discovered gate scripts binding a package root	8 (all `packages/spec`)
...of those, scripts carrying a hint dead at the repo root	2
dead-at-root hints in that population	6
...that re-anchor onto something the tree HAS	5
...refused because the re-anchored form reaches nothing	1 (`'zod'`, a package specifier)

⚠️The card's description of the shape does not survive the sweep, and this changed the rule. The card described the class as literals passed to resolve(PKG_DIR, …). That shape is real — 6 such literals across 6 of the 8 scripts, every one a tracked file — but it is not where the confirmed instance comes from. All five of check:generated's dead literals are artifact: values in its GATED ledger (artifact: 'api-surface/', artifact: 'src/meta-spelling/meta-url-data.generated.ts', …), describing what each gate verifies; nothing resolves them. A rule restricted to the syntactic shape would have repaired none of the three dead leads the card names. The restriction was measured and dropped; the reachability test is what keeps the wider reading honest.

The rule that landed

A hint is re-read against the package root its writer binds only when deepestTrackedPrefix is empty — no tracked path begins with even its first segment, which is exactly the branch whose reason text was false — and only if the re-anchored form reaches the tree through hintReachesTree. It replaces rather than adds. Neither priced refusal is relaxed and neither is reachable from here: admission is untouched (a literal that is not already a hint never arrives), and a re-anchoring that names nothing is dropped, so the rule cannot invent a path. The layout-moved class (a shorter surviving prefix) is deliberately left alone — its first segment is a real repo directory, so re-anchoring it would replace a triage lead with a fabricated one.

The price, before and after, on this base

Over 201 discovered families and 7900 tracked files:

	before	after
watch-hint (gate, file) pairs	142089	142139 (+50, 0 lost)
families whose hint set moves	—	1 (`check:generated`, no other)
pairs re-attributed (same pair, new via)	—	0 — every replaced spelling was dead, so it covered nothing to re-attribute
(check, hint) live / inert	1108 / 152	1113 / 147
distinct hints in the fleet	725	722 — three of the five re-anchored forms are already spelled from the root by other gates
families carrying dead literals	46	45
dead literals	152	147
...in the empty-`deepest` class	128	123

Provenance, which is the criterion this file prices and never volume: 5 of 5 admitted hints are TRUE leads, and 50 of 50 pairs with them, each verified at its declaration site — they are check:generated's own artifact ledger and verifying those artifacts is the gate's whole job:

re-anchored hint	files	ledger entry
`packages/spec/api-surface`	17	`check:api-surface`
`packages/spec/export-origins`	17	`check:export-origins`
`packages/spec/declaration-map`	14	`check:declaration-map`
`packages/spec/src/meta-spelling/meta-url-data.generated.ts`	1	`check:meta-url-spelling`
`packages/spec/liveness/state-counts.md`	1	the liveness counts

0 fabricated. For scale, hintCovers' docblock prices the bare-top-level-word admission it refuses at +139084 pairs on this corpus; this is 0.04% of it, and moduleRelativeDirectoryHint's naive-resolved-form widening was +53 hints for 1 true lead against this rule's 5 for 5.

The whole --residue output moves by exactly one row before and after (diffed line by line).

The reason text

The by-construction sentence ended never was a repo path — a claim about EVERY base made from evidence about ONE, and a dead-lead label printed over three live files. It now reads:

'HINT' — no tracked path under its first segment; a specifier, or a path anchored at a base this scan did not resolve

Checkable by the reader (git ls-files on the first segment is empty) and no longer asserting a base it did not test. The three docblocks stating the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep still lands on the incident. The pins move with the text: pinned to a retired phrase, the dist-freshness case would pass on any tree at all — a green over the exact regression it exists to catch.

Ablation

Anchor call replaced with return [...hints] on the committed tree, mutation proven on disk before the run (anchor occurrences 1 to 0, injected marker 1, blob 9743275c to e51f9d24), restore leg trapped and then proven by byte comparison (git diff HEAD empty, blob back to the HEAD blob 9743275c). Ablated, check:generated returns to 5 of 10 declared literal(s) reach nothing tracked with the three named literals marked. The ablated run also shows the corrected reason text in place, so the two halves are independent — the reason-text repair is the deliverable the card specified for the no-repair branch, and it stands on its own.

#14289 — prose only, no behaviour change

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM … and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the last key in coveringKey, an identity claim about a single FILE (#13000).

Re-derived at this PR's base, not copied from the card:

EDGE import: 27 families, 38 (family,target) pairs, 15 distinct targets
EDGE run: 1 family, 1 pair, 1 distinct target
EDGE manifest: 10 families, 10 pairs, 2 distinct targets
EDGE read: 12 families, 16 pairs, 15 distinct targets
INHERITED HINTS BY EDGE LABEL:
import: 373 hints across 13 families
manifest: 7 hints across 7 families
run: 1 hint across 1 family
read: 0

Read targets that carried a hint into their family: 0. Of the 16 read-edge pairs, 8 are answered by the READ identity key and 8 by an earlier key — consistent with coveringKey's "additive by construction" note.

The two numberings are reconciled rather than flattened: the docblock numbers all four scans in source order (so the manifest edge is the fourth), discoverFamilies numbers only the population follows (so the same edge is its third), and each now says which. The claim that there are three population follows all ending at a program is gone, and packageManifestTargets' "the other three follows take" is corrected to two.

Gates

Union re-run after the final commit, at git rev-parse --short HEAD = 70c1aedff (merge of origin/maind16df7418). Derived with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands — 15 families, all named by path, 0 by kind; the change set is exactly the one file. Exit codes captured to a file before any pipe.

gate	exit	verdict
`pnpm check:pm-dispatch-gates` (the tool's `--self-test`)	0	RAN-PASS — `dispatch-gates self-test: 1240 cases pass`
`node scripts/check-ci-filter-parity.mjs`	0	RAN-PASS
`node scripts/check-cross-package-test-inputs.mjs` / `pnpm check:cross-package-test-inputs`	0 / 0	RAN-PASS (both spellings)
`node scripts/check-self-test-wired.mjs`	0	RAN-PASS
`node scripts/check-shard-attestation.mjs`	0	RAN-PASS
`node scripts/check-test-completeness.mjs`	3	NOT MEASURED — no test log to hand it; the gate says so itself and calls exit 3 distinct from a finding's 1
`pnpm check:agent-test-spelling`	0	RAN-PASS
`pnpm check:bash32-floor`	0	RAN-PASS
`pnpm check:cli-command-ids`	0	RAN-PASS
`pnpm check:declared-population-live`	0	RAN-PASS
`pnpm check:entry-guard`	0	RAN-PASS
`pnpm check:parse-guard`	0	RAN-PASS
`pnpm check:pnpm-filter-targets`	0	RAN-PASS
`pnpm check:watch-hint-literal`	0	RAN-PASS
`pnpm check:ratchet-remedy-authority`	0	RAN-PASS (beyond the derived family)
`node scripts/check-self-test-workflow-commands.mjs`	0	RAN-PASS (beyond the derived family)
`pnpm check:nul-bytes`	0	RAN-PASS

Lint, narrowed and declared.pnpm lint is eslint . --no-inline-config over the whole repo; the run here is targeted at the one changed file: 1 file linted, 0 errors, 0 warnings, exit 0, the count read from --format json. The narrowing is a measurement rather than a gap because this repo runs one eslint.config.mjs which never enables type-aware linting for any file (no parserOptions.project, no typed rules — stated and positive-control-measured in that config's own header), so a one-file diff cannot move the verdict on any untouched file, and the file is a config match rather than an ignore (it produced a result object, not an ignore warning).

Changeset: none. This is a scripts/**-only diff that publishes nothing from any package, which is the repo convention on recent scripts-only PRs; the skip-changeset label carries it.

Generated by Claude Code

…iter binds extractWatchHints resolves a path literal against exactly two anchors: the repo root, and the writer's own directory when the author spelled a ./ or ../ prefix. A gate that lives inside a package binds a third at module top -- join(dirname(fileURLToPath(import.meta.url)), '..') -- and every literal it writes against that binding was read from the REPO root, where nothing of the sort exists. check:generated's five artifact-ledger literals are the confirmed instance: all five alive under packages/spec/ and all five printed by the residue as dead leads, under a reason text asserting they were never repo paths. Measured first, as the card required. Over the 201 discovered families' 195 scannable gate scripts, 8 bind a package root this way, 2 of them carry a hint dead at the repo root, 6 such hints exist, 5 re-anchor onto something the tree has and 1 is refused because it reaches nothing. Price: 142089 -> 142139 watch-hint (gate, file) pairs (+50, 0 lost, 0 re-attributed), one family's hint set moves, and 5 of 5 admitted hints are true leads verified at their declaration site. The admitted class is exactly the one whose residue sentence was false: a hint is re-anchored only when no tracked path begins with even its first segment. The layout-moved class is left to triage, admission is untouched, and a re-anchoring that reaches nothing is dropped, so the rule cannot invent a path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

unreachableReason's by-construction branch ended "never was a repo path" -- a claim about EVERY base, made from evidence about ONE. check:generated's api-surface, export-origins and src/meta-spelling/... all printed under it while sitting on disk under packages/spec/: a dead-lead label on a live path, telling the reader there is nothing to chase about three files they could open. The sentence now names what the sweep actually established (no tracked path under the literal's first segment) and the two ways that happens, without asserting the literal never existed anywhere. The three docblocks that stated the same universal as current fact are corrected with it; the two that quote it as history keep the quote and say it is retired, so a grep for it still lands on the incident. The pins move with the text rather than staying on the retired phrase: pinned to a string nothing renders, the dist-freshness case would pass on any tree at all -- a green over the exact regression it exists to catch -- so it now asserts against the sentence that branch prints today, and the sweep case asserts both halves. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

The fourth-edge docblock above packageManifestTargets opened "The three follows above all end at a PROGRAM ... and inherit that program's declared population". The middle one does not: readProgramTargetsInSource fills entry.reads / entry.readOrigin and nothing feeds those into hintsOfModule. The read edge reaches a card only through the LAST key in coveringKey, which is an identity claim about a single FILE and never a population. Re-derived at this PR's base: the read edge is 12 families, 16 (family,target) pairs over 15 distinct targets, and it carries 0 inherited hints -- against 373 for the import edge, 7 for the manifest edge and 1 for the run edge. Reading a program's TEXT is not performing its work, so identity-only is the correct behaviour and the sentence was the defect. No behaviour change. The two numberings inside the file also disagreed: discoverFamilies called the spawn follow the SECOND edge and the manifest follow the THIRD, while the docblock called the manifest one the FOURTH. They count different things -- source-order edges, and population follows -- so each now says which, and the claim that there are three population follows all ending at a program is gone. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WLJQhde67SeTccsmnBVarV

…spatch-gates-package-root-anchor

claude added 4 commits September 2, 2026 07:00

Merge remote-tracking branch 'origin/main' into claude/issue-14208-di…
70c1aed
…spatch-gates-package-root-anchor

github-actionsBot added the size/m label Sep 2, 2026

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Sep 2, 2026

baozhoutao marked this pull request as ready for review September 2, 2026 08:12

baozhoutao enabled auto-merge September 2, 2026 08:13

baozhoutao added this pull request to the merge queue Sep 2, 2026

Merged via the queue into main with commit 9e286e2Sep 2, 2026
36 checks passed

baozhoutao deleted the claude/issue-14208-dispatch-gates-package-root-anchor branch September 2, 2026 08:39

huangyiirene mentioned this pull request Sep 2, 2026
[finding] check:pm-dispatch-gates repeatedly holds the shared verify lock 500-900s+, starving the queue #14521
Closed

claude added 4 commits September 2, 2026 07:00

Merge remote-tracking branch 'origin/main' into claude/issue-14208-di…
70c1aed
…spatch-gates-package-root-anchor

github-actionsBot added the size/m label Sep 2, 2026

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Sep 2, 2026

baozhoutao marked this pull request as ready for review September 2, 2026 08:12

baozhoutao enabled auto-merge September 2, 2026 08:13

baozhoutao added this pull request to the merge queue Sep 2, 2026

Merged via the queue into main with commit 9e286e2Sep 2, 2026
36 checks passed

baozhoutao deleted the claude/issue-14208-dispatch-gates-package-root-anchor branch September 2, 2026 08:39

huangyiirene mentioned this pull request Sep 2, 2026
[finding] check:pm-dispatch-gates repeatedly holds the shared verify lock 500-900s+, starving the queue #14521
Closed

claude added 4 commits September 2, 2026 07:00

Merge remote-tracking branch 'origin/main' into claude/issue-14208-di…
70c1aed
…spatch-gates-package-root-anchor

github-actionsBot added the size/m label Sep 2, 2026

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Sep 2, 2026

baozhoutao marked this pull request as ready for review September 2, 2026 08:12

baozhoutao enabled auto-merge September 2, 2026 08:13

baozhoutao added this pull request to the merge queue Sep 2, 2026

Merged via the queue into main with commit 9e286e2Sep 2, 2026
36 checks passed

baozhoutao deleted the claude/issue-14208-dispatch-gates-package-root-anchor branch September 2, 2026 08:39

huangyiirene mentioned this pull request Sep 2, 2026
[finding] check:pm-dispatch-gates repeatedly holds the shared verify lock 500-900s+, starving the queue #14521
Closed

claude added 4 commits September 2, 2026 07:00

Merge remote-tracking branch 'origin/main' into claude/issue-14208-di…
70c1aed
…spatch-gates-package-root-anchor

github-actionsBot added the size/m label Sep 2, 2026

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Sep 2, 2026

baozhoutao marked this pull request as ready for review September 2, 2026 08:12

baozhoutao enabled auto-merge September 2, 2026 08:13

baozhoutao added this pull request to the merge queue Sep 2, 2026

Merged via the queue into main with commit 9e286e2Sep 2, 2026
36 checks passed

baozhoutao deleted the claude/issue-14208-dispatch-gates-package-root-anchor branch September 2, 2026 08:39

huangyiirene mentioned this pull request Sep 2, 2026
[finding] check:pm-dispatch-gates repeatedly holds the shared verify lock 500-900s+, starving the queue #14521
Closed

claude added 4 commits September 2, 2026 07:00

Merge remote-tracking branch 'origin/main' into claude/issue-14208-di…
70c1aed
…spatch-gates-package-root-anchor

github-actionsBot added the size/m label Sep 2, 2026

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Sep 2, 2026

baozhoutao marked this pull request as ready for review September 2, 2026 08:12

baozhoutao enabled auto-merge September 2, 2026 08:13

baozhoutao added this pull request to the merge queue Sep 2, 2026

Merged via the queue into main with commit 9e286e2Sep 2, 2026
36 checks passed

baozhoutao deleted the claude/issue-14208-dispatch-gates-package-root-anchor branch September 2, 2026 08:39

huangyiirene mentioned this pull request Sep 2, 2026
[finding] check:pm-dispatch-gates repeatedly holds the shared verify lock 500-900s+, starving the queue #14521
Closed

claude added 4 commits September 2, 2026 07:00

Merge remote-tracking branch 'origin/main' into claude/issue-14208-di…
70c1aed
…spatch-gates-package-root-anchor

github-actionsBot added the size/m label Sep 2, 2026

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Sep 2, 2026

baozhoutao marked this pull request as ready for review September 2, 2026 08:12

baozhoutao enabled auto-merge September 2, 2026 08:13

baozhoutao added this pull request to the merge queue Sep 2, 2026

Merged via the queue into main with commit 9e286e2Sep 2, 2026
36 checks passed

baozhoutao deleted the claude/issue-14208-dispatch-gates-package-root-anchor branch September 2, 2026 08:39

huangyiirene mentioned this pull request Sep 2, 2026
[finding] check:pm-dispatch-gates repeatedly holds the shared verify lock 500-900s+, starving the queue #14521
Closed

claude added 4 commits September 2, 2026 07:00

Merge remote-tracking branch 'origin/main' into claude/issue-14208-di…
70c1aed
…spatch-gates-package-root-anchor

github-actionsBot added the size/m label Sep 2, 2026

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Sep 2, 2026

baozhoutao marked this pull request as ready for review September 2, 2026 08:12

baozhoutao enabled auto-merge September 2, 2026 08:13

baozhoutao added this pull request to the merge queue Sep 2, 2026

Merged via the queue into main with commit 9e286e2Sep 2, 2026
36 checks passed

baozhoutao deleted the claude/issue-14208-dispatch-gates-package-root-anchor branch September 2, 2026 08:39

huangyiirene mentioned this pull request Sep 2, 2026
[finding] check:pm-dispatch-gates repeatedly holds the shared verify lock 500-900s+, starving the queue #14521
Closed

claude added 4 commits September 2, 2026 07:00

Merge remote-tracking branch 'origin/main' into claude/issue-14208-di…
70c1aed
…spatch-gates-package-root-anchor

github-actionsBot added the size/m label Sep 2, 2026

claudeBot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Sep 2, 2026

baozhoutao marked this pull request as ready for review September 2, 2026 08:12

baozhoutao enabled auto-merge September 2, 2026 08:13

baozhoutao added this pull request to the merge queue Sep 2, 2026

Merged via the queue into main with commit 9e286e2Sep 2, 2026
36 checks passed

baozhoutao deleted the claude/issue-14208-dispatch-gates-package-root-anchor branch September 2, 2026 08:39

huangyiirene mentioned this pull request Sep 2, 2026
[finding] check:pm-dispatch-gates repeatedly holds the shared verify lock 500-900s+, starving the queue #14521
Closed

Uh oh!

Conversation

claudeBot commented Sep 2, 2026

#14208 — measure first, as the card required

The population sweep

The rule that landed

The price, before and after, on this base

The reason text

Ablation

#14289 — prose only, no behaviour change

Gates

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

claudeBot commented Sep 2, 2026

#14208 — measure first, as the card required

The population sweep

The rule that landed

The price, before and after, on this base

The reason text

Ablation

#14289 — prose only, no behaviour change

Gates

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

claudeBot commented Sep 2, 2026

#14208 — measure first, as the card required

The population sweep

The rule that landed

The price, before and after, on this base

The reason text

Ablation

#14289 — prose only, no behaviour change

Gates

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

claudeBot commented Sep 2, 2026

#14208 — measure first, as the card required

The population sweep

The rule that landed

The price, before and after, on this base

The reason text

Ablation

#14289 — prose only, no behaviour change

Gates

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects