docs(pm-skill): a decision-box restore must cite a retrievable ruling id it fetched in the same pass

[claude]

Fixes#13730
Fixes#13792

⛔ Governed surface (.claude/**) — draft only. Do not flip ready, do not arm auto-merge, do not queue. Human merge, per Prime Directive #14.

What lands

Six lines in .claude/skills/pm-dispatch/references/dispatch-runbook.md, inside its ## 落卡与裁决记录细则 section, directly after the ruling-record atomic-four supplement they constrain. The rule:

A write that takes a card out ofneeds-user-decision must carry a comment id — a director-ledger entry or the card's own ruling comment — and that id must have been FETCHED in the same pass. A timestamp plus a recollection is not a citation, and the presence of an id is not retrievability. Reader side is symmetric: a decision re-read that hits a box exit with no retrievable id treats the card as unruled — same-write flip back, no dispatch — and the conclusion stays on verifiability, asserting no cause.

Two carriers are legal because the ruling of 2026-08-31 on the ruling-durability question made the director seat's per-session ledger table the record of a ruling's existence, with the card comment as elaboration. The rule text names both, so it does not fork from that ruling.

Why the criterion is "fetched", not "an id is present"

Both failure shapes were measured on this card's own evidence run, and a presence-only criterion passes both:

• A real id that is gone. Thirteen cards from the 2026-08-29 adjudication batch carry citations of their own ruling comment, written by a different account while the comment was still readable — several quoting it verbatim. Every one of those ids returns HTTP 404 today, in both repositories.
• An id that never existed. One seat wrote a ruling permalink into a clause-② line and then corrected itself: "That id is invented".

Fetching the id in the same pass is the one check that separates a ruling on record from a recollection, and it is machine-checkable in principle — deliberately not 「请仔细核对」, which is measured dead.

The second member (#13792) is delivered as a comment, not as a diff

The bound reconciliation deliverable — an 在案 / 不在案 list for the director ledger's 第 3 场 claim of 58 per-card ruling comments — is posted on #13792 as comment #13792 (comment). Summary of that reading, which is also the evidence base for the rule text above:

reading	result
the claimed batch, identified	58 decision-box exits, 2026-08-29T01:31:17Z–02:17:58Z, one actor, objectstack 35 + objectui 23, clustering into exactly 12 bursts of 5 = 批 #1–#12, containing all seven cards the ledger names
状态转换	58 / 58 在案 (label-event pairs, which survive comment deletion)
逐卡录裁评论	0 / 58 retrievable today
of those	13 曾在案、现不可取回 (cited by later comments, 404 now) · 45 with no citation at all
positive control, same query, later sessions	08-30: 19/30 and 14/15 · 08-31: 44/48 and 13/15 — so the zero is a reading, not a broken query

⛔ The ledger issue itself is not edited by this PR: the checkable list exists first, and the correction is the director seat's act on that evidence. ⛔ No cause is asserted for the missing comments.

Placement — and the option deliberately left open

references/state-machine.md is arguably the better home by readership (its own 出处 line says it is consulted at the moment of a transition, and a box exit is a transition). It is not used here, and the reason is measured rather than aesthetic: that file stands at 44 lines against a ceiling of 44, and so do decision-analysis.md (46/46) and lanes/director.md (77/77). Inserting the rule there turns check:pm-skill-ratchet red, and raising a ceiling requires a maintainer ruling quoted in the raising PR — a decision on the human floor, which this seat does not make.

⇒ Non-blocking, for the maintainer's discretion later: moving this rule to references/state-machine.md would cost a +N ceiling raise on that file. Nothing here depends on it; the runbook's ruling-record section is a correct home, and it is where the mechanical flag for "thread contains a ruling record but the card still carries needs-user-decision" already lives.

dispatch-runbook.md absorbs the six lines inside its existing budget: 271 lines against a ceiling of 278 (headroom 7 remains). No ratchet raise, no gate edit, no changeset (a .claude/**-only diff publishes nothing).

Verification

Gate union derived on this branch's own head — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no stale-tree warning — and every family run at head 6ddc73e:

EXIT=0 :: pnpm check:agent-test-spelling
EXIT=0 :: pnpm check:doc-authoring
EXIT=0 :: pnpm --filter @objectstack/lint run check:doc-formula-expressions
EXIT=0 :: pnpm check:pm-governed-merges
EXIT=0 :: pnpm check:pm-skill-id-lint
EXIT=0 :: pnpm check:pm-skill-ratchet
EXIT=0 :: pnpm check:skill-frame-sync

check:pm-skill-ratchet's own verdict line for the touched file: "references/dispatch-runbook.md is 271 lines (ceiling 278; headroom 7)". Two of those gates first reported PREREQUISITE NOT MET rather than a failure (@objectstack/formula and @objectstack/lint unbuilt in a fresh worktree); both were built and re-run, and the greens above are the post-build readings. Every new line measures at most 120 bytes, wrapped by the ratchet's own exported wrapLine.

Out of scope, recorded not expanded

• A patrol or gate that mechanizes the citation check — a follow-up if wanted, deliberately not built here.
• The ruling-durability design question one level up ([Decision] Is a maintainer ruling "recorded" when it exists only as a comment on a single card? — its durability is currently bounded by whoever happened to file that card #13766) is not answered here; it was ruled separately on 2026-08-31 and its skills-encoding item is queued on its own.
• The stall-takeover rule (skills: encode the review-chain-stall takeover rule (four conditions) into contract-review.md's clear-equals-land section #13791) is a different mechanism and is untouched.
• The account-scoped comment unreachability that explains the missing 2026-08-29 record is measured in the comment linked above and belongs to the already-open cards on that class ([finding] LIVE: os-elon is no longer assignable on this repo — every seat's claim protocol silently loses its assignee half, mid-session #13561, Dangling-reference patrol: nothing detects an issue reference that fails to RESOLVE (as distinct from closed) — measured 6+ times, and a card's author is unreadable once it is unreachable #13634, The tenant-audit census issue 13178 is unreachable (404) while three open cards cite it as their input #13548) — no new card filed, no cause asserted.

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#13730
Fixes#13792

⛔ Governed surface (.claude/**) — draft only. Do not flip ready, do not arm auto-merge, do not queue. Human merge, per Prime Directive #14.

What lands

Six lines in .claude/skills/pm-dispatch/references/dispatch-runbook.md, inside its ## 落卡与裁决记录细则 section, directly after the ruling-record atomic-four supplement they constrain. The rule:

A write that takes a card out ofneeds-user-decision must carry a comment id — a director-ledger entry or the card's own ruling comment — and that id must have been FETCHED in the same pass. A timestamp plus a recollection is not a citation, and the presence of an id is not retrievability. Reader side is symmetric: a decision re-read that hits a box exit with no retrievable id treats the card as unruled — same-write flip back, no dispatch — and the conclusion stays on verifiability, asserting no cause.

Two carriers are legal because the ruling of 2026-08-31 on the ruling-durability question made the director seat's per-session ledger table the record of a ruling's existence, with the card comment as elaboration. The rule text names both, so it does not fork from that ruling.

Why the criterion is "fetched", not "an id is present"

Both failure shapes were measured on this card's own evidence run, and a presence-only criterion passes both:

• A real id that is gone. Thirteen cards from the 2026-08-29 adjudication batch carry citations of their own ruling comment, written by a different account while the comment was still readable — several quoting it verbatim. Every one of those ids returns HTTP 404 today, in both repositories.
• An id that never existed. One seat wrote a ruling permalink into a clause-② line and then corrected itself: "That id is invented".

Fetching the id in the same pass is the one check that separates a ruling on record from a recollection, and it is machine-checkable in principle — deliberately not 「请仔细核对」, which is measured dead.

The second member (#13792) is delivered as a comment, not as a diff

The bound reconciliation deliverable — an 在案 / 不在案 list for the director ledger's 第 3 场 claim of 58 per-card ruling comments — is posted on #13792 as comment #13792 (comment). Summary of that reading, which is also the evidence base for the rule text above:

reading	result
the claimed batch, identified	58 decision-box exits, 2026-08-29T01:31:17Z–02:17:58Z, one actor, objectstack 35 + objectui 23, clustering into exactly 12 bursts of 5 = 批 #1–#12, containing all seven cards the ledger names
状态转换	58 / 58 在案 (label-event pairs, which survive comment deletion)
逐卡录裁评论	0 / 58 retrievable today
of those	13 曾在案、现不可取回 (cited by later comments, 404 now) · 45 with no citation at all
positive control, same query, later sessions	08-30: 19/30 and 14/15 · 08-31: 44/48 and 13/15 — so the zero is a reading, not a broken query

⛔ The ledger issue itself is not edited by this PR: the checkable list exists first, and the correction is the director seat's act on that evidence. ⛔ No cause is asserted for the missing comments.

Placement — and the option deliberately left open

references/state-machine.md is arguably the better home by readership (its own 出处 line says it is consulted at the moment of a transition, and a box exit is a transition). It is not used here, and the reason is measured rather than aesthetic: that file stands at 44 lines against a ceiling of 44, and so do decision-analysis.md (46/46) and lanes/director.md (77/77). Inserting the rule there turns check:pm-skill-ratchet red, and raising a ceiling requires a maintainer ruling quoted in the raising PR — a decision on the human floor, which this seat does not make.

⇒ Non-blocking, for the maintainer's discretion later: moving this rule to references/state-machine.md would cost a +N ceiling raise on that file. Nothing here depends on it; the runbook's ruling-record section is a correct home, and it is where the mechanical flag for "thread contains a ruling record but the card still carries needs-user-decision" already lives.

dispatch-runbook.md absorbs the six lines inside its existing budget: 271 lines against a ceiling of 278 (headroom 7 remains). No ratchet raise, no gate edit, no changeset (a .claude/**-only diff publishes nothing).

Verification

Gate union derived on this branch's own head — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no stale-tree warning — and every family run at head 6ddc73e:

EXIT=0 :: pnpm check:agent-test-spelling
EXIT=0 :: pnpm check:doc-authoring
EXIT=0 :: pnpm --filter @objectstack/lint run check:doc-formula-expressions
EXIT=0 :: pnpm check:pm-governed-merges
EXIT=0 :: pnpm check:pm-skill-id-lint
EXIT=0 :: pnpm check:pm-skill-ratchet
EXIT=0 :: pnpm check:skill-frame-sync

check:pm-skill-ratchet's own verdict line for the touched file: "references/dispatch-runbook.md is 271 lines (ceiling 278; headroom 7)". Two of those gates first reported PREREQUISITE NOT MET rather than a failure (@objectstack/formula and @objectstack/lint unbuilt in a fresh worktree); both were built and re-run, and the greens above are the post-build readings. Every new line measures at most 120 bytes, wrapped by the ratchet's own exported wrapLine.

Out of scope, recorded not expanded

• A patrol or gate that mechanizes the citation check — a follow-up if wanted, deliberately not built here.
• The ruling-durability design question one level up ([Decision] Is a maintainer ruling "recorded" when it exists only as a comment on a single card? — its durability is currently bounded by whoever happened to file that card #13766) is not answered here; it was ruled separately on 2026-08-31 and its skills-encoding item is queued on its own.
• The stall-takeover rule (skills: encode the review-chain-stall takeover rule (four conditions) into contract-review.md's clear-equals-land section #13791) is a different mechanism and is untouched.
• The account-scoped comment unreachability that explains the missing 2026-08-29 record is measured in the comment linked above and belongs to the already-open cards on that class ([finding] LIVE: os-elon is no longer assignable on this repo — every seat's claim protocol silently loses its assignee half, mid-session #13561, Dangling-reference patrol: nothing detects an issue reference that fails to RESOLVE (as distinct from closed) — measured 6+ times, and a card's author is unreadable once it is unreachable #13634, The tenant-audit census issue 13178 is unreachable (404) while three open cards cite it as their input #13548) — no new card filed, no cause asserted.

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#13730
Fixes#13792

⛔ Governed surface (.claude/**) — draft only. Do not flip ready, do not arm auto-merge, do not queue. Human merge, per Prime Directive #14.

What lands

Six lines in .claude/skills/pm-dispatch/references/dispatch-runbook.md, inside its ## 落卡与裁决记录细则 section, directly after the ruling-record atomic-four supplement they constrain. The rule:

A write that takes a card out ofneeds-user-decision must carry a comment id — a director-ledger entry or the card's own ruling comment — and that id must have been FETCHED in the same pass. A timestamp plus a recollection is not a citation, and the presence of an id is not retrievability. Reader side is symmetric: a decision re-read that hits a box exit with no retrievable id treats the card as unruled — same-write flip back, no dispatch — and the conclusion stays on verifiability, asserting no cause.

Two carriers are legal because the ruling of 2026-08-31 on the ruling-durability question made the director seat's per-session ledger table the record of a ruling's existence, with the card comment as elaboration. The rule text names both, so it does not fork from that ruling.

Why the criterion is "fetched", not "an id is present"

Both failure shapes were measured on this card's own evidence run, and a presence-only criterion passes both:

• A real id that is gone. Thirteen cards from the 2026-08-29 adjudication batch carry citations of their own ruling comment, written by a different account while the comment was still readable — several quoting it verbatim. Every one of those ids returns HTTP 404 today, in both repositories.
• An id that never existed. One seat wrote a ruling permalink into a clause-② line and then corrected itself: "That id is invented".

Fetching the id in the same pass is the one check that separates a ruling on record from a recollection, and it is machine-checkable in principle — deliberately not 「请仔细核对」, which is measured dead.

The second member (#13792) is delivered as a comment, not as a diff

The bound reconciliation deliverable — an 在案 / 不在案 list for the director ledger's 第 3 场 claim of 58 per-card ruling comments — is posted on #13792 as comment #13792 (comment). Summary of that reading, which is also the evidence base for the rule text above:

reading	result
the claimed batch, identified	58 decision-box exits, 2026-08-29T01:31:17Z–02:17:58Z, one actor, objectstack 35 + objectui 23, clustering into exactly 12 bursts of 5 = 批 #1–#12, containing all seven cards the ledger names
状态转换	58 / 58 在案 (label-event pairs, which survive comment deletion)
逐卡录裁评论	0 / 58 retrievable today
of those	13 曾在案、现不可取回 (cited by later comments, 404 now) · 45 with no citation at all
positive control, same query, later sessions	08-30: 19/30 and 14/15 · 08-31: 44/48 and 13/15 — so the zero is a reading, not a broken query

⛔ The ledger issue itself is not edited by this PR: the checkable list exists first, and the correction is the director seat's act on that evidence. ⛔ No cause is asserted for the missing comments.

Placement — and the option deliberately left open

references/state-machine.md is arguably the better home by readership (its own 出处 line says it is consulted at the moment of a transition, and a box exit is a transition). It is not used here, and the reason is measured rather than aesthetic: that file stands at 44 lines against a ceiling of 44, and so do decision-analysis.md (46/46) and lanes/director.md (77/77). Inserting the rule there turns check:pm-skill-ratchet red, and raising a ceiling requires a maintainer ruling quoted in the raising PR — a decision on the human floor, which this seat does not make.

⇒ Non-blocking, for the maintainer's discretion later: moving this rule to references/state-machine.md would cost a +N ceiling raise on that file. Nothing here depends on it; the runbook's ruling-record section is a correct home, and it is where the mechanical flag for "thread contains a ruling record but the card still carries needs-user-decision" already lives.

dispatch-runbook.md absorbs the six lines inside its existing budget: 271 lines against a ceiling of 278 (headroom 7 remains). No ratchet raise, no gate edit, no changeset (a .claude/**-only diff publishes nothing).

Verification

Gate union derived on this branch's own head — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no stale-tree warning — and every family run at head 6ddc73e:

EXIT=0 :: pnpm check:agent-test-spelling
EXIT=0 :: pnpm check:doc-authoring
EXIT=0 :: pnpm --filter @objectstack/lint run check:doc-formula-expressions
EXIT=0 :: pnpm check:pm-governed-merges
EXIT=0 :: pnpm check:pm-skill-id-lint
EXIT=0 :: pnpm check:pm-skill-ratchet
EXIT=0 :: pnpm check:skill-frame-sync

check:pm-skill-ratchet's own verdict line for the touched file: "references/dispatch-runbook.md is 271 lines (ceiling 278; headroom 7)". Two of those gates first reported PREREQUISITE NOT MET rather than a failure (@objectstack/formula and @objectstack/lint unbuilt in a fresh worktree); both were built and re-run, and the greens above are the post-build readings. Every new line measures at most 120 bytes, wrapped by the ratchet's own exported wrapLine.

Out of scope, recorded not expanded

• A patrol or gate that mechanizes the citation check — a follow-up if wanted, deliberately not built here.
• The ruling-durability design question one level up ([Decision] Is a maintainer ruling "recorded" when it exists only as a comment on a single card? — its durability is currently bounded by whoever happened to file that card #13766) is not answered here; it was ruled separately on 2026-08-31 and its skills-encoding item is queued on its own.
• The stall-takeover rule (skills: encode the review-chain-stall takeover rule (four conditions) into contract-review.md's clear-equals-land section #13791) is a different mechanism and is untouched.
• The account-scoped comment unreachability that explains the missing 2026-08-29 record is measured in the comment linked above and belongs to the already-open cards on that class ([finding] LIVE: os-elon is no longer assignable on this repo — every seat's claim protocol silently loses its assignee half, mid-session #13561, Dangling-reference patrol: nothing detects an issue reference that fails to RESOLVE (as distinct from closed) — measured 6+ times, and a card's author is unreadable once it is unreachable #13634, The tenant-audit census issue 13178 is unreachable (404) while three open cards cite it as their input #13548) — no new card filed, no cause asserted.

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#13730
Fixes#13792

⛔ Governed surface (.claude/**) — draft only. Do not flip ready, do not arm auto-merge, do not queue. Human merge, per Prime Directive #14.

What lands

Six lines in .claude/skills/pm-dispatch/references/dispatch-runbook.md, inside its ## 落卡与裁决记录细则 section, directly after the ruling-record atomic-four supplement they constrain. The rule:

A write that takes a card out ofneeds-user-decision must carry a comment id — a director-ledger entry or the card's own ruling comment — and that id must have been FETCHED in the same pass. A timestamp plus a recollection is not a citation, and the presence of an id is not retrievability. Reader side is symmetric: a decision re-read that hits a box exit with no retrievable id treats the card as unruled — same-write flip back, no dispatch — and the conclusion stays on verifiability, asserting no cause.

Two carriers are legal because the ruling of 2026-08-31 on the ruling-durability question made the director seat's per-session ledger table the record of a ruling's existence, with the card comment as elaboration. The rule text names both, so it does not fork from that ruling.

Why the criterion is "fetched", not "an id is present"

Both failure shapes were measured on this card's own evidence run, and a presence-only criterion passes both:

• A real id that is gone. Thirteen cards from the 2026-08-29 adjudication batch carry citations of their own ruling comment, written by a different account while the comment was still readable — several quoting it verbatim. Every one of those ids returns HTTP 404 today, in both repositories.
• An id that never existed. One seat wrote a ruling permalink into a clause-② line and then corrected itself: "That id is invented".

Fetching the id in the same pass is the one check that separates a ruling on record from a recollection, and it is machine-checkable in principle — deliberately not 「请仔细核对」, which is measured dead.

The second member (#13792) is delivered as a comment, not as a diff

The bound reconciliation deliverable — an 在案 / 不在案 list for the director ledger's 第 3 场 claim of 58 per-card ruling comments — is posted on #13792 as comment #13792 (comment). Summary of that reading, which is also the evidence base for the rule text above:

reading	result
the claimed batch, identified	58 decision-box exits, 2026-08-29T01:31:17Z–02:17:58Z, one actor, objectstack 35 + objectui 23, clustering into exactly 12 bursts of 5 = 批 #1–#12, containing all seven cards the ledger names
状态转换	58 / 58 在案 (label-event pairs, which survive comment deletion)
逐卡录裁评论	0 / 58 retrievable today
of those	13 曾在案、现不可取回 (cited by later comments, 404 now) · 45 with no citation at all
positive control, same query, later sessions	08-30: 19/30 and 14/15 · 08-31: 44/48 and 13/15 — so the zero is a reading, not a broken query

⛔ The ledger issue itself is not edited by this PR: the checkable list exists first, and the correction is the director seat's act on that evidence. ⛔ No cause is asserted for the missing comments.

Placement — and the option deliberately left open

references/state-machine.md is arguably the better home by readership (its own 出处 line says it is consulted at the moment of a transition, and a box exit is a transition). It is not used here, and the reason is measured rather than aesthetic: that file stands at 44 lines against a ceiling of 44, and so do decision-analysis.md (46/46) and lanes/director.md (77/77). Inserting the rule there turns check:pm-skill-ratchet red, and raising a ceiling requires a maintainer ruling quoted in the raising PR — a decision on the human floor, which this seat does not make.

⇒ Non-blocking, for the maintainer's discretion later: moving this rule to references/state-machine.md would cost a +N ceiling raise on that file. Nothing here depends on it; the runbook's ruling-record section is a correct home, and it is where the mechanical flag for "thread contains a ruling record but the card still carries needs-user-decision" already lives.

dispatch-runbook.md absorbs the six lines inside its existing budget: 271 lines against a ceiling of 278 (headroom 7 remains). No ratchet raise, no gate edit, no changeset (a .claude/**-only diff publishes nothing).

Verification

Gate union derived on this branch's own head — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no stale-tree warning — and every family run at head 6ddc73e:

EXIT=0 :: pnpm check:agent-test-spelling
EXIT=0 :: pnpm check:doc-authoring
EXIT=0 :: pnpm --filter @objectstack/lint run check:doc-formula-expressions
EXIT=0 :: pnpm check:pm-governed-merges
EXIT=0 :: pnpm check:pm-skill-id-lint
EXIT=0 :: pnpm check:pm-skill-ratchet
EXIT=0 :: pnpm check:skill-frame-sync

check:pm-skill-ratchet's own verdict line for the touched file: "references/dispatch-runbook.md is 271 lines (ceiling 278; headroom 7)". Two of those gates first reported PREREQUISITE NOT MET rather than a failure (@objectstack/formula and @objectstack/lint unbuilt in a fresh worktree); both were built and re-run, and the greens above are the post-build readings. Every new line measures at most 120 bytes, wrapped by the ratchet's own exported wrapLine.

Out of scope, recorded not expanded

• A patrol or gate that mechanizes the citation check — a follow-up if wanted, deliberately not built here.
• The ruling-durability design question one level up ([Decision] Is a maintainer ruling "recorded" when it exists only as a comment on a single card? — its durability is currently bounded by whoever happened to file that card #13766) is not answered here; it was ruled separately on 2026-08-31 and its skills-encoding item is queued on its own.
• The stall-takeover rule (skills: encode the review-chain-stall takeover rule (four conditions) into contract-review.md's clear-equals-land section #13791) is a different mechanism and is untouched.
• The account-scoped comment unreachability that explains the missing 2026-08-29 record is measured in the comment linked above and belongs to the already-open cards on that class ([finding] LIVE: os-elon is no longer assignable on this repo — every seat's claim protocol silently loses its assignee half, mid-session #13561, Dangling-reference patrol: nothing detects an issue reference that fails to RESOLVE (as distinct from closed) — measured 6+ times, and a card's author is unreadable once it is unreachable #13634, The tenant-audit census issue 13178 is unreachable (404) while three open cards cite it as their input #13548) — no new card filed, no cause asserted.

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#13730
Fixes#13792

⛔ Governed surface (.claude/**) — draft only. Do not flip ready, do not arm auto-merge, do not queue. Human merge, per Prime Directive #14.

What lands

Six lines in .claude/skills/pm-dispatch/references/dispatch-runbook.md, inside its ## 落卡与裁决记录细则 section, directly after the ruling-record atomic-four supplement they constrain. The rule:

A write that takes a card out ofneeds-user-decision must carry a comment id — a director-ledger entry or the card's own ruling comment — and that id must have been FETCHED in the same pass. A timestamp plus a recollection is not a citation, and the presence of an id is not retrievability. Reader side is symmetric: a decision re-read that hits a box exit with no retrievable id treats the card as unruled — same-write flip back, no dispatch — and the conclusion stays on verifiability, asserting no cause.

Two carriers are legal because the ruling of 2026-08-31 on the ruling-durability question made the director seat's per-session ledger table the record of a ruling's existence, with the card comment as elaboration. The rule text names both, so it does not fork from that ruling.

Why the criterion is "fetched", not "an id is present"

Both failure shapes were measured on this card's own evidence run, and a presence-only criterion passes both:

• A real id that is gone. Thirteen cards from the 2026-08-29 adjudication batch carry citations of their own ruling comment, written by a different account while the comment was still readable — several quoting it verbatim. Every one of those ids returns HTTP 404 today, in both repositories.
• An id that never existed. One seat wrote a ruling permalink into a clause-② line and then corrected itself: "That id is invented".

Fetching the id in the same pass is the one check that separates a ruling on record from a recollection, and it is machine-checkable in principle — deliberately not 「请仔细核对」, which is measured dead.

The second member (#13792) is delivered as a comment, not as a diff

The bound reconciliation deliverable — an 在案 / 不在案 list for the director ledger's 第 3 场 claim of 58 per-card ruling comments — is posted on #13792 as comment #13792 (comment). Summary of that reading, which is also the evidence base for the rule text above:

reading	result
the claimed batch, identified	58 decision-box exits, 2026-08-29T01:31:17Z–02:17:58Z, one actor, objectstack 35 + objectui 23, clustering into exactly 12 bursts of 5 = 批 #1–#12, containing all seven cards the ledger names
状态转换	58 / 58 在案 (label-event pairs, which survive comment deletion)
逐卡录裁评论	0 / 58 retrievable today
of those	13 曾在案、现不可取回 (cited by later comments, 404 now) · 45 with no citation at all
positive control, same query, later sessions	08-30: 19/30 and 14/15 · 08-31: 44/48 and 13/15 — so the zero is a reading, not a broken query

⛔ The ledger issue itself is not edited by this PR: the checkable list exists first, and the correction is the director seat's act on that evidence. ⛔ No cause is asserted for the missing comments.

Placement — and the option deliberately left open

references/state-machine.md is arguably the better home by readership (its own 出处 line says it is consulted at the moment of a transition, and a box exit is a transition). It is not used here, and the reason is measured rather than aesthetic: that file stands at 44 lines against a ceiling of 44, and so do decision-analysis.md (46/46) and lanes/director.md (77/77). Inserting the rule there turns check:pm-skill-ratchet red, and raising a ceiling requires a maintainer ruling quoted in the raising PR — a decision on the human floor, which this seat does not make.

⇒ Non-blocking, for the maintainer's discretion later: moving this rule to references/state-machine.md would cost a +N ceiling raise on that file. Nothing here depends on it; the runbook's ruling-record section is a correct home, and it is where the mechanical flag for "thread contains a ruling record but the card still carries needs-user-decision" already lives.

dispatch-runbook.md absorbs the six lines inside its existing budget: 271 lines against a ceiling of 278 (headroom 7 remains). No ratchet raise, no gate edit, no changeset (a .claude/**-only diff publishes nothing).

Verification

Gate union derived on this branch's own head — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no stale-tree warning — and every family run at head 6ddc73e:

EXIT=0 :: pnpm check:agent-test-spelling
EXIT=0 :: pnpm check:doc-authoring
EXIT=0 :: pnpm --filter @objectstack/lint run check:doc-formula-expressions
EXIT=0 :: pnpm check:pm-governed-merges
EXIT=0 :: pnpm check:pm-skill-id-lint
EXIT=0 :: pnpm check:pm-skill-ratchet
EXIT=0 :: pnpm check:skill-frame-sync

check:pm-skill-ratchet's own verdict line for the touched file: "references/dispatch-runbook.md is 271 lines (ceiling 278; headroom 7)". Two of those gates first reported PREREQUISITE NOT MET rather than a failure (@objectstack/formula and @objectstack/lint unbuilt in a fresh worktree); both were built and re-run, and the greens above are the post-build readings. Every new line measures at most 120 bytes, wrapped by the ratchet's own exported wrapLine.

Out of scope, recorded not expanded

• A patrol or gate that mechanizes the citation check — a follow-up if wanted, deliberately not built here.
• The ruling-durability design question one level up ([Decision] Is a maintainer ruling "recorded" when it exists only as a comment on a single card? — its durability is currently bounded by whoever happened to file that card #13766) is not answered here; it was ruled separately on 2026-08-31 and its skills-encoding item is queued on its own.
• The stall-takeover rule (skills: encode the review-chain-stall takeover rule (four conditions) into contract-review.md's clear-equals-land section #13791) is a different mechanism and is untouched.
• The account-scoped comment unreachability that explains the missing 2026-08-29 record is measured in the comment linked above and belongs to the already-open cards on that class ([finding] LIVE: os-elon is no longer assignable on this repo — every seat's claim protocol silently loses its assignee half, mid-session #13561, Dangling-reference patrol: nothing detects an issue reference that fails to RESOLVE (as distinct from closed) — measured 6+ times, and a card's author is unreadable once it is unreachable #13634, The tenant-audit census issue 13178 is unreachable (404) while three open cards cite it as their input #13548) — no new card filed, no cause asserted.

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#13730
Fixes#13792

⛔ Governed surface (.claude/**) — draft only. Do not flip ready, do not arm auto-merge, do not queue. Human merge, per Prime Directive #14.

What lands

Six lines in .claude/skills/pm-dispatch/references/dispatch-runbook.md, inside its ## 落卡与裁决记录细则 section, directly after the ruling-record atomic-four supplement they constrain. The rule:

A write that takes a card out ofneeds-user-decision must carry a comment id — a director-ledger entry or the card's own ruling comment — and that id must have been FETCHED in the same pass. A timestamp plus a recollection is not a citation, and the presence of an id is not retrievability. Reader side is symmetric: a decision re-read that hits a box exit with no retrievable id treats the card as unruled — same-write flip back, no dispatch — and the conclusion stays on verifiability, asserting no cause.

Two carriers are legal because the ruling of 2026-08-31 on the ruling-durability question made the director seat's per-session ledger table the record of a ruling's existence, with the card comment as elaboration. The rule text names both, so it does not fork from that ruling.

Why the criterion is "fetched", not "an id is present"

Both failure shapes were measured on this card's own evidence run, and a presence-only criterion passes both:

• A real id that is gone. Thirteen cards from the 2026-08-29 adjudication batch carry citations of their own ruling comment, written by a different account while the comment was still readable — several quoting it verbatim. Every one of those ids returns HTTP 404 today, in both repositories.
• An id that never existed. One seat wrote a ruling permalink into a clause-② line and then corrected itself: "That id is invented".

Fetching the id in the same pass is the one check that separates a ruling on record from a recollection, and it is machine-checkable in principle — deliberately not 「请仔细核对」, which is measured dead.

The second member (#13792) is delivered as a comment, not as a diff

The bound reconciliation deliverable — an 在案 / 不在案 list for the director ledger's 第 3 场 claim of 58 per-card ruling comments — is posted on #13792 as comment #13792 (comment). Summary of that reading, which is also the evidence base for the rule text above:

reading	result
the claimed batch, identified	58 decision-box exits, 2026-08-29T01:31:17Z–02:17:58Z, one actor, objectstack 35 + objectui 23, clustering into exactly 12 bursts of 5 = 批 #1–#12, containing all seven cards the ledger names
状态转换	58 / 58 在案 (label-event pairs, which survive comment deletion)
逐卡录裁评论	0 / 58 retrievable today
of those	13 曾在案、现不可取回 (cited by later comments, 404 now) · 45 with no citation at all
positive control, same query, later sessions	08-30: 19/30 and 14/15 · 08-31: 44/48 and 13/15 — so the zero is a reading, not a broken query

⛔ The ledger issue itself is not edited by this PR: the checkable list exists first, and the correction is the director seat's act on that evidence. ⛔ No cause is asserted for the missing comments.

Placement — and the option deliberately left open

references/state-machine.md is arguably the better home by readership (its own 出处 line says it is consulted at the moment of a transition, and a box exit is a transition). It is not used here, and the reason is measured rather than aesthetic: that file stands at 44 lines against a ceiling of 44, and so do decision-analysis.md (46/46) and lanes/director.md (77/77). Inserting the rule there turns check:pm-skill-ratchet red, and raising a ceiling requires a maintainer ruling quoted in the raising PR — a decision on the human floor, which this seat does not make.

⇒ Non-blocking, for the maintainer's discretion later: moving this rule to references/state-machine.md would cost a +N ceiling raise on that file. Nothing here depends on it; the runbook's ruling-record section is a correct home, and it is where the mechanical flag for "thread contains a ruling record but the card still carries needs-user-decision" already lives.

dispatch-runbook.md absorbs the six lines inside its existing budget: 271 lines against a ceiling of 278 (headroom 7 remains). No ratchet raise, no gate edit, no changeset (a .claude/**-only diff publishes nothing).

Verification

Gate union derived on this branch's own head — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no stale-tree warning — and every family run at head 6ddc73e:

EXIT=0 :: pnpm check:agent-test-spelling
EXIT=0 :: pnpm check:doc-authoring
EXIT=0 :: pnpm --filter @objectstack/lint run check:doc-formula-expressions
EXIT=0 :: pnpm check:pm-governed-merges
EXIT=0 :: pnpm check:pm-skill-id-lint
EXIT=0 :: pnpm check:pm-skill-ratchet
EXIT=0 :: pnpm check:skill-frame-sync

check:pm-skill-ratchet's own verdict line for the touched file: "references/dispatch-runbook.md is 271 lines (ceiling 278; headroom 7)". Two of those gates first reported PREREQUISITE NOT MET rather than a failure (@objectstack/formula and @objectstack/lint unbuilt in a fresh worktree); both were built and re-run, and the greens above are the post-build readings. Every new line measures at most 120 bytes, wrapped by the ratchet's own exported wrapLine.

Out of scope, recorded not expanded

• A patrol or gate that mechanizes the citation check — a follow-up if wanted, deliberately not built here.
• The ruling-durability design question one level up ([Decision] Is a maintainer ruling "recorded" when it exists only as a comment on a single card? — its durability is currently bounded by whoever happened to file that card #13766) is not answered here; it was ruled separately on 2026-08-31 and its skills-encoding item is queued on its own.
• The stall-takeover rule (skills: encode the review-chain-stall takeover rule (four conditions) into contract-review.md's clear-equals-land section #13791) is a different mechanism and is untouched.
• The account-scoped comment unreachability that explains the missing 2026-08-29 record is measured in the comment linked above and belongs to the already-open cards on that class ([finding] LIVE: os-elon is no longer assignable on this repo — every seat's claim protocol silently loses its assignee half, mid-session #13561, Dangling-reference patrol: nothing detects an issue reference that fails to RESOLVE (as distinct from closed) — measured 6+ times, and a card's author is unreadable once it is unreachable #13634, The tenant-audit census issue 13178 is unreachable (404) while three open cards cite it as their input #13548) — no new card filed, no cause asserted.

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#13730
Fixes#13792

⛔ Governed surface (.claude/**) — draft only. Do not flip ready, do not arm auto-merge, do not queue. Human merge, per Prime Directive #14.

What lands

Six lines in .claude/skills/pm-dispatch/references/dispatch-runbook.md, inside its ## 落卡与裁决记录细则 section, directly after the ruling-record atomic-four supplement they constrain. The rule:

A write that takes a card out ofneeds-user-decision must carry a comment id — a director-ledger entry or the card's own ruling comment — and that id must have been FETCHED in the same pass. A timestamp plus a recollection is not a citation, and the presence of an id is not retrievability. Reader side is symmetric: a decision re-read that hits a box exit with no retrievable id treats the card as unruled — same-write flip back, no dispatch — and the conclusion stays on verifiability, asserting no cause.

Two carriers are legal because the ruling of 2026-08-31 on the ruling-durability question made the director seat's per-session ledger table the record of a ruling's existence, with the card comment as elaboration. The rule text names both, so it does not fork from that ruling.

Why the criterion is "fetched", not "an id is present"

Both failure shapes were measured on this card's own evidence run, and a presence-only criterion passes both:

• A real id that is gone. Thirteen cards from the 2026-08-29 adjudication batch carry citations of their own ruling comment, written by a different account while the comment was still readable — several quoting it verbatim. Every one of those ids returns HTTP 404 today, in both repositories.
• An id that never existed. One seat wrote a ruling permalink into a clause-② line and then corrected itself: "That id is invented".

Fetching the id in the same pass is the one check that separates a ruling on record from a recollection, and it is machine-checkable in principle — deliberately not 「请仔细核对」, which is measured dead.

The second member (#13792) is delivered as a comment, not as a diff

The bound reconciliation deliverable — an 在案 / 不在案 list for the director ledger's 第 3 场 claim of 58 per-card ruling comments — is posted on #13792 as comment #13792 (comment). Summary of that reading, which is also the evidence base for the rule text above:

reading	result
the claimed batch, identified	58 decision-box exits, 2026-08-29T01:31:17Z–02:17:58Z, one actor, objectstack 35 + objectui 23, clustering into exactly 12 bursts of 5 = 批 #1–#12, containing all seven cards the ledger names
状态转换	58 / 58 在案 (label-event pairs, which survive comment deletion)
逐卡录裁评论	0 / 58 retrievable today
of those	13 曾在案、现不可取回 (cited by later comments, 404 now) · 45 with no citation at all
positive control, same query, later sessions	08-30: 19/30 and 14/15 · 08-31: 44/48 and 13/15 — so the zero is a reading, not a broken query

⛔ The ledger issue itself is not edited by this PR: the checkable list exists first, and the correction is the director seat's act on that evidence. ⛔ No cause is asserted for the missing comments.

Placement — and the option deliberately left open

references/state-machine.md is arguably the better home by readership (its own 出处 line says it is consulted at the moment of a transition, and a box exit is a transition). It is not used here, and the reason is measured rather than aesthetic: that file stands at 44 lines against a ceiling of 44, and so do decision-analysis.md (46/46) and lanes/director.md (77/77). Inserting the rule there turns check:pm-skill-ratchet red, and raising a ceiling requires a maintainer ruling quoted in the raising PR — a decision on the human floor, which this seat does not make.

⇒ Non-blocking, for the maintainer's discretion later: moving this rule to references/state-machine.md would cost a +N ceiling raise on that file. Nothing here depends on it; the runbook's ruling-record section is a correct home, and it is where the mechanical flag for "thread contains a ruling record but the card still carries needs-user-decision" already lives.

dispatch-runbook.md absorbs the six lines inside its existing budget: 271 lines against a ceiling of 278 (headroom 7 remains). No ratchet raise, no gate edit, no changeset (a .claude/**-only diff publishes nothing).

Verification

Gate union derived on this branch's own head — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no stale-tree warning — and every family run at head 6ddc73e:

EXIT=0 :: pnpm check:agent-test-spelling
EXIT=0 :: pnpm check:doc-authoring
EXIT=0 :: pnpm --filter @objectstack/lint run check:doc-formula-expressions
EXIT=0 :: pnpm check:pm-governed-merges
EXIT=0 :: pnpm check:pm-skill-id-lint
EXIT=0 :: pnpm check:pm-skill-ratchet
EXIT=0 :: pnpm check:skill-frame-sync

check:pm-skill-ratchet's own verdict line for the touched file: "references/dispatch-runbook.md is 271 lines (ceiling 278; headroom 7)". Two of those gates first reported PREREQUISITE NOT MET rather than a failure (@objectstack/formula and @objectstack/lint unbuilt in a fresh worktree); both were built and re-run, and the greens above are the post-build readings. Every new line measures at most 120 bytes, wrapped by the ratchet's own exported wrapLine.

Out of scope, recorded not expanded

• A patrol or gate that mechanizes the citation check — a follow-up if wanted, deliberately not built here.
• The ruling-durability design question one level up ([Decision] Is a maintainer ruling "recorded" when it exists only as a comment on a single card? — its durability is currently bounded by whoever happened to file that card #13766) is not answered here; it was ruled separately on 2026-08-31 and its skills-encoding item is queued on its own.
• The stall-takeover rule (skills: encode the review-chain-stall takeover rule (four conditions) into contract-review.md's clear-equals-land section #13791) is a different mechanism and is untouched.
• The account-scoped comment unreachability that explains the missing 2026-08-29 record is measured in the comment linked above and belongs to the already-open cards on that class ([finding] LIVE: os-elon is no longer assignable on this repo — every seat's claim protocol silently loses its assignee half, mid-session #13561, Dangling-reference patrol: nothing detects an issue reference that fails to RESOLVE (as distinct from closed) — measured 6+ times, and a card's author is unreadable once it is unreachable #13634, The tenant-audit census issue 13178 is unreachable (404) while three open cards cite it as their input #13548) — no new card filed, no cause asserted.

Generated by Claude Code

---

Generated by Claude Code

[claude]

Fixes#13730
Fixes#13792

⛔ Governed surface (.claude/**) — draft only. Do not flip ready, do not arm auto-merge, do not queue. Human merge, per Prime Directive #14.

What lands

Six lines in .claude/skills/pm-dispatch/references/dispatch-runbook.md, inside its ## 落卡与裁决记录细则 section, directly after the ruling-record atomic-four supplement they constrain. The rule:

A write that takes a card out ofneeds-user-decision must carry a comment id — a director-ledger entry or the card's own ruling comment — and that id must have been FETCHED in the same pass. A timestamp plus a recollection is not a citation, and the presence of an id is not retrievability. Reader side is symmetric: a decision re-read that hits a box exit with no retrievable id treats the card as unruled — same-write flip back, no dispatch — and the conclusion stays on verifiability, asserting no cause.

Two carriers are legal because the ruling of 2026-08-31 on the ruling-durability question made the director seat's per-session ledger table the record of a ruling's existence, with the card comment as elaboration. The rule text names both, so it does not fork from that ruling.

Why the criterion is "fetched", not "an id is present"

Both failure shapes were measured on this card's own evidence run, and a presence-only criterion passes both:

• A real id that is gone. Thirteen cards from the 2026-08-29 adjudication batch carry citations of their own ruling comment, written by a different account while the comment was still readable — several quoting it verbatim. Every one of those ids returns HTTP 404 today, in both repositories.
• An id that never existed. One seat wrote a ruling permalink into a clause-② line and then corrected itself: "That id is invented".

Fetching the id in the same pass is the one check that separates a ruling on record from a recollection, and it is machine-checkable in principle — deliberately not 「请仔细核对」, which is measured dead.

The second member (#13792) is delivered as a comment, not as a diff

The bound reconciliation deliverable — an 在案 / 不在案 list for the director ledger's 第 3 场 claim of 58 per-card ruling comments — is posted on #13792 as comment #13792 (comment). Summary of that reading, which is also the evidence base for the rule text above:

reading	result
the claimed batch, identified	58 decision-box exits, 2026-08-29T01:31:17Z–02:17:58Z, one actor, objectstack 35 + objectui 23, clustering into exactly 12 bursts of 5 = 批 #1–#12, containing all seven cards the ledger names
状态转换	58 / 58 在案 (label-event pairs, which survive comment deletion)
逐卡录裁评论	0 / 58 retrievable today
of those	13 曾在案、现不可取回 (cited by later comments, 404 now) · 45 with no citation at all
positive control, same query, later sessions	08-30: 19/30 and 14/15 · 08-31: 44/48 and 13/15 — so the zero is a reading, not a broken query

⛔ The ledger issue itself is not edited by this PR: the checkable list exists first, and the correction is the director seat's act on that evidence. ⛔ No cause is asserted for the missing comments.

Placement — and the option deliberately left open

references/state-machine.md is arguably the better home by readership (its own 出处 line says it is consulted at the moment of a transition, and a box exit is a transition). It is not used here, and the reason is measured rather than aesthetic: that file stands at 44 lines against a ceiling of 44, and so do decision-analysis.md (46/46) and lanes/director.md (77/77). Inserting the rule there turns check:pm-skill-ratchet red, and raising a ceiling requires a maintainer ruling quoted in the raising PR — a decision on the human floor, which this seat does not make.

⇒ Non-blocking, for the maintainer's discretion later: moving this rule to references/state-machine.md would cost a +N ceiling raise on that file. Nothing here depends on it; the runbook's ruling-record section is a correct home, and it is where the mechanical flag for "thread contains a ruling record but the card still carries needs-user-decision" already lives.

dispatch-runbook.md absorbs the six lines inside its existing budget: 271 lines against a ceiling of 278 (headroom 7 remains). No ratchet raise, no gate edit, no changeset (a .claude/**-only diff publishes nothing).

Verification

Gate union derived on this branch's own head — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, no stale-tree warning — and every family run at head 6ddc73e:

EXIT=0 :: pnpm check:agent-test-spelling
EXIT=0 :: pnpm check:doc-authoring
EXIT=0 :: pnpm --filter @objectstack/lint run check:doc-formula-expressions
EXIT=0 :: pnpm check:pm-governed-merges
EXIT=0 :: pnpm check:pm-skill-id-lint
EXIT=0 :: pnpm check:pm-skill-ratchet
EXIT=0 :: pnpm check:skill-frame-sync

check:pm-skill-ratchet's own verdict line for the touched file: "references/dispatch-runbook.md is 271 lines (ceiling 278; headroom 7)". Two of those gates first reported PREREQUISITE NOT MET rather than a failure (@objectstack/formula and @objectstack/lint unbuilt in a fresh worktree); both were built and re-run, and the greens above are the post-build readings. Every new line measures at most 120 bytes, wrapped by the ratchet's own exported wrapLine.

Out of scope, recorded not expanded

• A patrol or gate that mechanizes the citation check — a follow-up if wanted, deliberately not built here.
• The ruling-durability design question one level up ([Decision] Is a maintainer ruling "recorded" when it exists only as a comment on a single card? — its durability is currently bounded by whoever happened to file that card #13766) is not answered here; it was ruled separately on 2026-08-31 and its skills-encoding item is queued on its own.
• The stall-takeover rule (skills: encode the review-chain-stall takeover rule (four conditions) into contract-review.md's clear-equals-land section #13791) is a different mechanism and is untouched.
• The account-scoped comment unreachability that explains the missing 2026-08-29 record is measured in the comment linked above and belongs to the already-open cards on that class ([finding] LIVE: os-elon is no longer assignable on this repo — every seat's claim protocol silently loses its assignee half, mid-session #13561, Dangling-reference patrol: nothing detects an issue reference that fails to RESOLVE (as distinct from closed) — measured 6+ times, and a card's author is unreadable once it is unreachable #13634, The tenant-audit census issue 13178 is unreachable (404) while three open cards cite it as their input #13548) — no new card filed, no cause asserted.

Generated by Claude Code

---

Generated by Claude Code