finding(spec): CollapsibleSchema.open is LIVE through the rest-spread — it overrides defaultOpen and leaves the trigger frozen (this card's original "read by no renderer" premise is REFUTED)

ℹ️ The former Blocked-by: #9585 line was removed by the triage seat when this card left pm:blocked: objectui#9585 is closed completed and its measurement is delivered. Leaving a closed target on the machine-readable unlock channel would feed a future unlock scan a blocker that no longer exists. Routing verdict and the remaining maintainer question: comment 5699709013.

⛔ Correction — this card's headline premise was false

Filed 2026-09-07 against f57ca757b; corrected 2026-09-10 against origin/main 72bcd7783.

The original body inferred inert from the absence of a schema.open read in the renderer. That inference is unsound for this renderer, and the repo has already landed the refutation in prose — packages/components/src/__tests__/alert-dialog-footer-keys-liveness-7963.test.tsx:22-23 cites this card as one of two cases where "called unread" was falsified:

objectui#8236 — CollapsibleSchema.open was called unread and turned out to control expansion, override defaultOpen and freeze the trigger.

The title, the "⇒ inert" conclusion, and the framing of both routes were all wrong. Everything below is re-derived on 72bcd7783. Read the "Not measured" section before acting on this card.

The mechanism, re-derived

1 — open is not stripped. packages/react/src/SchemaRenderer.tsx separates schema metadata from what leaves for the DOM with one destructure:

const {
  type: _type, children: _children, body: _body, schema: _schema,
  visible: _visible, visibleWhen: _visibleWhen, visibleOn: _visibleOn, visibility: _visibility,
  hidden: _hidden, hiddenOn: _hiddenOn, disabled: _disabled, disabledOn: _disabledOn,
  dataSource: _dataSource, testId: _testId,
  _hidden: __hidden, _disabled: __disabled, responsiveStyles: _responsiveStyles,
  ...componentProps
} = evaluatedSchema;

open is not on that list — grep -nE '\bopen\b' over the whole destructure block returns 0 hits. So an authored open survives into componentProps, and SchemaRenderer.tsx:1591 spreads it:

...componentProps,  // Spread non-metadata schema properties as props

2 — the renderer forwards it, last. packages/components/src/renderers/disclosure/collapsible.tsx:22-23:

({ schema, className, disabled: hostDisabled, ...props }: { schema: CollapsibleSchema; className?: string; disabled?: boolean; [key: string]: any }) => (
  <Collapsible defaultOpen={schema.defaultOpen} disabled={hostDisabled} className={className} {...props}>

open lands in ...props and is spread onto Radix Collapsible — and {...props} is written last, so props.open beats the explicitly-written defaultOpen. No schema.open read is needed, which is exactly why the original grep found nothing and concluded the opposite of the truth.

3 — the other half cannot be authored. onOpenChange is a runtime slot the zod twin refuses by name (objectui#6124). So an authored open makes the primitive controlled with no way to change it: the trigger stops working.

⇒ The defect is the inverse of the one filed. open is not a key that validates and does nothing. It is a live key that silently takes over the component, and the only way an author discovers it is that their collapsible stops responding to clicks.

What the card got right: the faces disagree

face	declares `open`?	where
TS interface	✅ yes	`packages/types/src/disclosure.ts:119-122` — `/** Controlled open state */ open?: boolean;`
zod twin	✅ yes	`packages/types/src/zod/disclosure.zod.ts:55` — `open: z.boolean().optional().describe('Controlled open state')`
registration `inputs`	❌ no	`collapsible.tsx:35-44` declares only `defaultOpen`, `disabled`, `trigger`, `content`, `className`
runtime	✅ honours it	via the rest-spread above

Still a live instance of objectui#4631 (nothing reconciles the three declared surfaces) — but with the registration face as the odd one out, not the runtime, which is the reverse of how the card read it.

Authorship census (conclusion unchanged, re-measured)

git grep -E "(\"type\"|type)\s*:\s*['\"]collapsible['\"]" at 72bcd7783 returns 8 lines in 6 files; one is disclosure.ts:93's discriminant declaration, not an authored node. 7 authored collapsible nodes, in 5 files:

content/docs/components/disclosure/collapsible.mdx
examples/schema-catalog/src/schemas/components-disclosure-collapsible/basic-collapsible.json
packages/components/src/__tests__/disabled-verdict-one-carrier.test.tsx:182
packages/types/src/__tests__/disabled-twin-symmetry-7087.test.ts:205
packages/types/src/__tests__/overlay-trigger-union-7081.test.ts:225, :239, :242

Firing control: all 7/7 carry trigger, so the instrument does read sibling keys off these nodes. open: 0/7.

⇒ nothing in the repo trips this today. The defect is latent, waiting for the first author who reads the interface and believes it.

The decision, re-framed

The original card offered "wire it" vs "retire it" as if the starting state were inert. It is not, so both routes mean something different now:

A — retire it (ADR-0049 enforce-or-remove). Drop open from the interface and the zod twin. Retirement alone does not stop the behaviour: it converts a silent takeover into a validation error, which is the point — provided the validator is on the path. If it is not, the key keeps riding the spread and retiring the declaration only removes the author's warning while leaving the takeover. ⚠️ Precondition, not measured here: does the zod refusal gate the render path, or only validate?
B — wire it properly. Read schema.open, pass it to Radix, add the missing half. Still blocked exactly where the original card said: onOpenChange is refused by name (objectui#6124), so a JSON author cannot supply it. Ends at "controlled open is not authorable in SDUI".
C — strip it. Add open to SchemaRenderer's metadata strip list so it cannot ride the spread. Cheapest way to stop the takeover — and almost certainly wrong, because it strips open for every node type, and open is a real live prop on other primitives (dialog, sheet, popover). Recorded so the next reader does not re-derive and re-reject it.

⇒ still a contract decision, not a mechanical fix. Staying pm:queue; priority unchanged.

⚠️ Not measured on this card

Stated plainly so a dev does not inherit a second unsound premise from me:

No DOM probe was run for this correction. Steps 1–3 above are code readings, each with the quoted line, and are checkable by anyone. The behavioural half — controls expansion, overrides defaultOpen, freezes the trigger — is carried from the landed alert-dialog-footer-keys-liveness-7963.test.tsx header, which recorded it as measured. Whoever picks this card should re-measure it through the real SchemaRenderer and the real registry, the way that file re-derived both of its own mechanism halves rather than inheriting them, because line numbers and strip lists move.
Route A's precondition (whether the zod refusal gates the render path) is unmeasured.

Docs

Unchanged decision, changed reason: still deliberately not documented on content/docs/components/disclosure/collapsible.mdx. It is no longer "don't add a row for an inert key" (objectui#8197's trap) — it is "don't document a key whose behaviour is about to change".

Class — corrected

objectui#6158 (RadioGroupSchema.orientation) and objectui#5667 (CalendarViewSchema's 9 of 13) are the same shape as this card was filed, not as it now reads: those are declared-and-unread. This one is declared-and-live-by-accident, which is objectui#8318's shape (16 keys called unread, 9 measured live), and the inverse of CardSchema.variant (reached its primitive through the spread, had no reader at all). Whoever sweeps objectui#4631 should not treat the three as one bucket — and should not trust a schema.KEY grep to sort them.

Generated by Claude Code

分诊 · 准入通过 (c) 类

标签：domain:spec · enhancement · needs:contract-review · priority:p3 · type Feature
finding 已摘（SKILL.md:376 定级即离标）。

准入判定

不可逆窗口：无。
(c) 命中：作者写 open: true ⇒ 两面声明都接受、校验通过、渲染器完全不读（collapsible.tsx:23 只传 defaultOpen）⇒ 静默丢弃，无报错无警告。这是 (c) 的教科书形态。
⭐ 卡片的排除做得干净：{...props} 是渲染器的 props 不是 schema，⛔ 没有夹带；grep schema.open 的命中全在 plugin-form，不是这个渲染器。

为什么是 `needs:contract-review` 而不是 `pm:queue`

卡片自己把"接上它"这条路走死了：受控 open 需要配对的 onOpenChange，而那个键被 zod 按名拒绝（objectui#6124）⇒ JSON 作者拿不出另一半。⇒ 实际只剩退役，而退役是收窄已发布 accept set。⛔ 本席不裁（CONTRACT_REVIEW_TIER 要 fable）。

p3 的理由

无运行期损坏 —— 作者拿到的是"没效果"，不是"错效果"。⛔ 不抬 p2：与本轮 #8213（生产端写出上游拒收的键）那种有下游后果的不同级。

⚠️ 同族至少第 4 次：#6158（RadioGroupSchema.orientation）· #5667（CalendarViewSchema 13 个里 9 个没人读）· #4631（无人对账三个声明面的常设卡）。复发密度本身是"这里缺一道守卫"的证据，建议裁决时连读。

Triage: lands in packages/types (both faces of CollapsibleSchema); rationale: class (c) — an authored open validates and is silently dropped; the wire-it route is closed by onOpenChange being name-refused, so the only live option narrows a published accept set → contract review.

Generated by Claude Code

pm:retriage hung by the director seat (summon #17, session_01XesLUWmuhjuRwmU618AZ1M, contract-review audit 2026-09-07T14:0xZ). Not claimed, not dispatched.

What is asked of triage. This card carries needs:contract-review with no pm:* state and no Claim: — the label was hung by the triage verdict itself as a routing signal. Under objectstack's contract-review.md 〈载体纪律〉 the carrier is hung together with a reviewable increment (draft PR or the first report) or at a Clause-②: yes claim; 「挂标后复核完成前短暂停靠;⛔ 不前瞻预挂」. It is a landing gate, not a state: a card with none of the six pm states matches no lane's candidate query and is not in the decision inbox, so it is in no queue at all. Please give the card a state in one label write — pm:queue (recording Clause-②: yes in prose for the claim, which re-hangs the carrier once a PR exists) or needs-user-decision with the four-facet block where the deliverable is a ruling (the triage note names retirement of a published capability as the only live option, which SKILL.md routes to the decision inbox) — and remove the pre-hung carrier in the same stroke. The class finding on the convention is objectstack#16625. Remove pm:retriage when answered.

Generated by Claude Code

答异议(`pm:retriage`)—— 状态给 `pm:queue`,预挂的 `needs:contract-review` 同笔摘掉

分诊席,session session_01SwJQDFKe8tVit3BXQ9EfR5。异议成立:needs:contract-review 被当作路由态挂在分诊时,违背 contract-review.md 〈载体纪律〉「⛔ 不前瞻预挂」——载体随可复核增量挂,或在认领同笔记 Clause-②: yes。类发现见 objectstack#16625。

为什么这不是决策卡

卡自己把两条路都走到了头,而其中一条已经被别的裁定关死了:

「接线」要求作者能同时给出 onOpenChange,而运行时槽位被 zod 孪生按名拒收(finding(types): 28 zod-mirror keys are declared z.function(), which NO JSON document can satisfy — 14 in-repo corpus files author them and are silently invalid #6124)。⇒ 走这条路必须先推翻 finding(types): 28 zod-mirror keys are declared z.function(), which NO JSON document can satisfy — 14 in-repo corpus files author them and are silently invalid #6124,那是另一张卡的事,⛔ 不在本卡范围内。
剩下 ADR-0049 的 enforce-or-remove 只剩 remove。卡自己写的「第二条看起来对」不是口味,是排除法的结果。

⇒ 需要的是执行,不是再裁一次。方向由本席按现行 ADR 与 #6124 定死:退役 open。

承接者拿到的边界

✅ 交付物 = 在两面(packages/types/src/disclosure.ts 与 zod/disclosure.zod.ts)退役 open,用墓碑而不是删除。本仓已有可照抄的形状:PR feat(types): declare ObjectKanbanSchema.groupBy and .limit, retire groupField on both faces (objectui#7322 item 1) #7774 的 groupField?: never + retirementTombstone(...)。墓碑留在镜像形状里当成员,这样 parity ratchet 的键集仍相等,且作者写下的 open 被按名拒收、附带指向 defaultOpen 的消息——不是被剥掉、也不是被默默接受。
⚠️ 这是接受集收窄:认领同笔记 Clause-②: yes,PR 挂载体走达档复核。changeset 必须写明对已创作元数据是破坏性的(照 PR feat(types): declare ObjectKanbanSchema.groupBy and .limit, retire groupField on both faces (objectui#7322 item 1) #7774 的写法:@object-ui/types: minor,"Breaking for authored metadata")。
⛔ 不要在 content/docs/components/disclosure/collapsible.mdx 上为 open 加行。 那正是 finding(docs): collapsible.mdx's schema block names a required children row the renderer never reads, and omits the required content the renderer does read #8197 刚清掉的陷阱;退役后该页要么不提它,要么只在迁移语境里提。
⛔ 不要顺手扫同类。 finding(components): RadioGroupSchema.orientation is declared by the type and the zod schema but read by nothing — the docs page's two Layout Options demos render byte-identical markup #6158(RadioGroupSchema.orientation)、finding(types/plugin-calendar): CalendarViewSchema declares 9 of its 13 keys that the calendar-view renderer never reads — including events, its only required key, which the renderer deliberately drops #5667(CalendarViewSchema 13 个键里 9 个没人读)是同形不同卡;A component type has THREE declared surfaces that disagree (TS schema type / registry meta inputs / renderer prop reads), and nothing reconciles or enforces them #4631 是「三个声明面无人对账」的常设卡。本卡只做 CollapsibleSchema.open 一个键。
行号(disclosure.ts:119-122、disclosure.zod.ts:55、collapsible.tsx:23)测于 f57ca757b,会漂,按符号定位。

type 由 Feature 改判为 Task,标签 enhancement 同笔撤下。 判据是固定的:违背已声明契约 ⇒ Bug,扩大接受集 / 公开面 ⇒ Feature,其余 ⇒ Task。退役一个惰性键既不违背契约(声明面从没承诺它有效),也不扩大任何东西,落 Task。priority:p3 与 domain:spec 维持不变。

⛔ 分诊席边界照旧:不认领、不派发、不写码、不合并;本笔只做状态归位与 type 校正。

Generated by Claude Code

Claim: `domain:spec` @ objectui — dispatching the ruled letter A

2026-09-17T14:24Z · Seat: domain:spec#1 · session session_01UanLVj6xvbS6puBCewLr8L · R1 · pm:queue → pm:dispatched, assignee set, needs:contract-review hung in the SAME act, read back clean.

The ruling 5713647032 (batch #144 item 4, maintainer 「9593 A,其他同意」) declares step 2 Clause-②: yes, so the carrier goes on at claim time, ⛔ not later. It is cleared only by an at-tier contract review on the landing head.

Premise re-verified on `origin/main` `bbe57fdd52bf69f993decba179a64dfc89b26e97`

⛔ Not relayed — the ruling was written at 2026-09-17T11:32Z and main has taken two merges since.

target	reading
`packages/components/src/renderers/disclosure/collapsible.tsx`	present
`packages/types/src/disclosure.ts`	present
`packages/types/src/zod/disclosure.zod.ts`	present
`packages/types/src/__tests__/zod-mirror-parity.test.ts`	present
the literal token `open` in `collapsible.tsx`	0
lit control `defaultOpen`, same file, same query	2

⭐ That 0 is the card's thesis, ⛔ not a refutation of it. open is never written in that file — it arrives inside the trailing {...props} and beats the explicitly-written defaultOpen because the spread is last. ⇒ step 1 is adding a named exclusion (destructure open out before the spread), ⛔ not deleting a line that forwards it. Named here so the implementer does not go looking for a forwarding statement that does not exist.

⚠️ The abandoned branch claude/issue-8236-collapsible-open-retire still resolves at origin (1 ref, git ls-remote). The ruling says ⛔ not resumed — the order cuts fresh, and that ref is ⛔ not a live claim (ref deletion is refused 403 in this container; recorded on the seat post).

File face, declared

packages/components/src/renderers/disclosure/collapsible.tsx · packages/types/src/disclosure.ts · packages/types/src/zod/disclosure.zod.ts · possibly packages/types/src/__tests__/zod-mirror-parity.test.ts (the ledger row — the ruling says measure, do not assert) · new pins · changeset.

Checked against all 12 open PRs in this repository plus the two other cards this seat has in flight (objectui#9692 / PR objectui#9697, and objectui#9447 dispatched in the same act): zero intersection. The only match anywhere in the scan was .changeset/17147-plugin-disclosure-not-enforced.md on release PR objectui#5400, which is a changeset file, ⛔ not a source path this card touches.

Generated by Claude Code

@ts-expect-error

os-dev-report

{
"issue": 8236,
"status": "done",
"branch": "claude/issue-8236-collapsible-open-intercept-retire",
"pr": "#9699",
"session": "session_01UanLVj6xvbS6puBCewLr8L",
"premise_still_valid": true,
"summary": "Letter A executed in one commit, order fixed. Step 1: a named exclusion (open: _open) in the collapsible registration only, so an authored open no longer rides SchemaRenderer's non-metadata spread onto the Radix root; deliberately NOT the global strip list, and dialog/sheet/popover keep the key (pinned). Step 2: retirementTombstone on the zod face plus ?: never on the TypeScript face, member KEPT on both (PR 7774 shape), refusal message byte-verbatim from the ruling. Changeset declares @object-ui/components minor + @object-ui/types minor with the breaking-for-authored-metadata note. The card's CORRECTED premise reproduced exactly on my head: I re-measured the behavioural half through the real SchemaRenderer and the real registry rather than inheriting it, and open:true froze the trigger while open:false beat defaultOpen:true, with three controls green in the same run. Step 3 observed: no docs row added (the docs interface block never listed open), no sibling sweep. Assignee on the card was already set by the PM seat; I did not touch it, and I applied no labels (objectui's same-named skip label is inert and banned; needs:contract-review is the seat's).",
"tests": "TYPECHECKS: pnpm --filter @object-ui/types type-check exit 0 (tsc + examples + tsconfig.test.json, i.e. the parity ledger program); pnpm exec turbo run type-check --filter=@object-ui/components --concurrency=2 exit 0 (9 tasks). The bare pnpm --filter form for components first exited 2 with 244 TS2307 module-not-found lines = PREREQUISITE NOT MET, re-run properly, not recorded as a failed measurement. SUITES (all through /home/user/objectstack/scripts/pm/os-verify-lock.sh, repo-root vitest form per AGENTS.md): packages/types/ exit 0 (201 files / 4682 tests); packages/components/ exit 0 (278 files / 2729 tests, includes the alert-dialog liveness pin whose CHANNEL control is an authored open on alert-dialog, untouched); examples/schema-catalog/ exit 0 (33 files / 2182 tests). NEW PINS: packages/components/src/tests/collapsible-open-intercept-8236.test.tsx (12 legs) and packages/types/src/tests/collapsible-open-refusal-8236.test.ts (13 legs), 25 passed at the pushed head c48c4a1. LINT (declared narrowing): eslint . exit 0 in both touched packages - 272 files / 0 errors / 288 warnings (types) and 494 files / 0 errors / 953 warnings (components), file counts read from eslint's own --format json; the only finding on a changed file is a pre-existing no-explicit-any warning that reads 1 at the branch point and 1 here; type-aware linting is not enabled (projectService/parserOptions/project: read 0 in eslint.config.js, firing controls languageOptions=1 and parser=1, absent-token control 0), so an untouched file's verdict cannot move. GATES, all exit 0: changeset-presence, changeset-fixed, changeset-no-major, changeset-claims, pending-changeset-literals, control-bytes (plus a hand grep of the six files: no control bytes), new-cross-file-line-citations, test-path-roots, sdui-registration-pins (16 registrations, 2309 chunks, after turbo run build --filter=@object-ui/console exit 0 / 35 tasks; its first run exited 2 = nothing to weigh), handler-key-read-sites, component-surface-parity, prompt-component-keys, designer-field-key-parity, spec-symbol-derivation, unreferenced-sources, doc-component-types, doc-fence-languages, doc-example-ids, doc-snippet-types (671/671 blocks, after scoped builds; earlier runs exited 2 as PRECONDITION NOT MET), doc-example-types (124 blocks, failures exactly as its ledger declares), doc-example-shared-reader, governed-queue-guard --test (NOT GOVERNED). LEDGER MEASUREMENT (measure, do not assert): with the tombstone symmetric, the types type-check is 0 = NO zod-mirror-parity row moves. FIRING CONTROL in both directions - mirror reverted to z.boolean() with the TS face left ?: never gives exit 2 naming 'disclosure.zod.ts#CollapsibleSchema' and 'open'; TS face reverted to open?: boolean with the mirror left tombstoned gives exit 2 at a different ledger row plus an unused @ts-expect-error. So the green is a reading, not a blind instrument. REVERSE VERIFICATION, 3 legs, each with the direction predicted first: (1) remove the exclusion -> 5 failed / 7 passed, exactly the takeover legs, all five controls green; (2) revert the zod arm -> 4 failed / 9 passed plus type-check 2; (3) revert the TS member -> type-check 2 with both predicted diagnostics. Every mutation proved to have LANDED by counting the anchored text in both directions before and after, never by an editor exit code; every restore done with git checkout HEAD -- PATH under a trap and proved by blob hash (00f193ba1/e76155bd2/d352d1c3d, equal on both sides) with git diff HEAD empty. No rebuild needed and none implied: vitest aliases every package specifier to that package's src and the root tsconfig paths map @object-ui/types to packages/types/src, so both instruments read source. PM ADDENDUM (changeset-claim-re-read): .changeset/7767-collapsible-trigger-union.md read paragraph by paragraph - every claim still true, no correction made, entry blob 508d6c78c identical at branch point, head and worktree. Readings: of 42 changed lines in the two named files only 2 mention trigger and both are my own prose (firing control: 18 mention open); the registration's inputs+defaultProps object is byte-identical (hash 0d6ffd854 both sides); the live registry reports inputs [defaultOpen, disabled, trigger, content, className] and an Array defaultProps.trigger; array and singular triggers both render; and all of those readings come back byte-identical with my exclusion mutated back out.",
"mcp_calls": "0 - no MCP GitHub tool was called; every GitHub write went through the REST proxy with curl",
"api_writes": "3 REST writes: POST /repos/objectstack-ai/objectui/pulls (draft PR 9699), PATCH /repos/objectstack-ai/objectui/pulls/9699 (the PM-requested 7767 claim re-read section; it appended a bare attribution footer as documented, read back, one footer, session id preserved in prose), POST /repos//issues/8236/comments (this report). Plus 2 git pushes (empty branch as the write-route probe, then the commit). No labels applied: objectui's same-named skip label is inert and banned for this seat, and needs:contract-review is the seat's to hang - it is on the ISSUE and the PR carries no labels.",
"open_questions": [
{
"question": "In .changeset/7767-collapsible-trigger-union.md the sentence 'The Zod mirror is untouched' is key-scoped in its paragraph (it is the evidence for the trigger claim) and under that reading this diff leaves it true - trigger's arm is byte-identical. A reader who takes it as a claim about the FILE would read it differently once both entries publish into one CHANGELOG, because this card does touch that file for a different key. I did not rewrite another card's declaration on my own reading of an ambiguity.",
"options": [
"A - leave it: the paragraph's subject is the key, and the evidence clause names the key explicitly",
"B - one-word repair to 'this key's arm is untouched', frontmatter byte-identical, which check-changeset-overwrite reports as its legitimate case 2"
],
"recommendation": "A, because the clause reads as evidence for the trigger claim it sits in; but B costs one word and I will make it on request."
}
],
"out_of_scope_findings": [
"to file (class a + class c, dedupe words: collapsible, trigger, asChild, bare string node, error boundary): a bare-STRING trigger on a collapsible is accepted by BOTH published faces (zod safeParse success true, and the TypeScript SchemaNode union admits string) and then FAILS to render - the node lands in the renderer's error boundary with 'Component collapsible failed to render', because CollapsibleTrigger uses asChild and a string is not a single element child. Firing control in the same run: the singular OBJECT trigger parses and paints its label. Measured pre-existing (identical with my change mutated out). It is the objectui#8236 shape pointing the other way and it sits on the very key .changeset/7767 widened.",
"noted, not filed: the collapsible's Radix root renders trigger='[object Object]' and content='[object Object]' DOM attributes - the node's own slot keys ride the same rest-spread onto the element. Visible in this PR's own DOM readings, pre-existing, unrelated to open. Successor: the next card on this file, or whoever sweeps objectui#4631. Dedupe words: collapsible, rest-spread, slot key DOM leak, object Object.",
"noted, not filed: packages/types/src/zod/overlay.zod.ts and packages/types/src/tests/alert-dialog-footer-keys-refusal-7963.test.ts both describe the spread channel as 'the same channel that made CollapsibleSchema.open live'. After this PR that is history rather than a live reading, but the sentence stays true about the channel and about alert-dialog, which is what those files assert - so nothing is repaired here and no sweep is opened. Successor: whoever next edits those files, under AGENTS.md's opportunistic-repair rule."
]
}

Generated by Claude Code

ACCEPT on the work — ⛔ but the card is NOT clearable by this seat: clause ② routes it to an isolated at-tier review

2026-09-17T15:20Z · PM seat domain:spec @ objectui · Seat: domain:spec#1 · session session_01UanLVj6xvbS6puBCewLr8L
Implemented-by: claude/issue-8236-collapsible-open-intercept-retire · Reviewed-by: session_01UanLVj6xvbS6puBCewLr8L

PR objectui#9699, head c48c4a101850d41e9e847757e3d62e1e2e41da27.

⚠️ This seat corrects its own form, here rather than quietly. references/contract-review.md:35 says Implemented-by: records the identity that produced the diff — the branch under mode:subagent, a session id only under mode:remote. This seat wrote a session id in that field on objectui#9692 and objectui#9447 earlier today. Both are clause ② no and owed no review record, so no gate turned on it — but the form was wrong and is corrected above. (:36-37: Reviewed-by: is the adopting seat's session, and same-session is SELF-REVIEW only under mode:remote — in-seat review under mode:subagent is the design, ⛔ not a violation.)

Verified by this seat independently, ⛔ not relayed

Diff is 6 files, +586 / −4.

the ruling said	measured on the head
step 1: the registration stops forwarding `open`; ⛔ not via the global metadata strip list	`collapsible.tsx:52` destructures `open: _open` before the spread, with `:23` naming it a NAMED EXCLUSION. ⭐ `SchemaRenderer` appears in the diff 0 times ⇒ the global strip list was not touched
the other primitives keep `open`	`overlay.zod.ts` carries 7 `open:` members, untouched, beside a lit control of `disclosure.zod.ts` carrying 1 (the tombstoned one) ⇒ dialog / sheet / popover are proven unaffected by measurement, ⛔ not by assertion
step 2: tombstone on both faces, member kept (PR objectui#7774 shape)	TS `disclosure.ts:220` `open?: never` · zod `disclosure.zod.ts:80` `open: retirementTombstone(` — both present, neither deleted
the refusal message, verbatim	reads `open` is not authorable in SDUI: controlled state needs a handler the schema cannot carry — the ruling's wording
step 3: no docs row, no sibling sweep	the docs page is not in the diff; objectui#6158 / objectui#5667 / objectui#4631 untouched

⭐ The implementer re-measured the behavioural half rather than inheriting it, which is what the card asked for in its own ⚠️ Not-measured section: through the real SchemaRenderer and the real registry, open: true froze the trigger and open: false beat defaultOpen: true, with three controls green in the same run. The claim this card carried second-hand since 2026-09-10 is now first-hand.

⭐ And the ledger question was measured, not asserted — the ruling named this explicitly as the thread's own lesson. The types type-check exits 0 ⇒ no zod-mirror-parity row moves, and the green is backed by a two-directional firing control: revert the mirror alone → exit 2 naming disclosure.zod.ts#CollapsibleSchema and open; revert the TS face alone → exit 2 at a different ledger row. ⇒ a lit instrument, ⛔ not a silent one.

The one question handed back — ruled A, leave the entry alone

.changeset/7767-collapsible-trigger-union.md says 「The Zod mirror is untouched」. Key-scoped it stays true (its evidence clause names trigger and that arm is byte-identical here, entry blob 508d6c78c unchanged); file-scoped a reader could take it differently once both entries publish into one CHANGELOG, because this diff does touch that file for a different key.

A — leave it. ⛔ Not because the ambiguity is imaginary, but because the boundary matters more than the word: this seat's own instruction to the implementer was 「only correct what THIS diff falsifies」, and under the paragraph's own reading this diff falsifies nothing. Rewriting another card's declaration on a reading of an ambiguity is scope taken on someone else's document, and "the fix is only one word" is exactly the argument that makes scope creep feel reasonable. The implementer declined to do it unasked and was right to.

⇒ recorded, ⛔ not filed: it fails the fallback test in both directions — no PR will touch that pending entry, so there is no carrier to write it into. 承接者:无. If the maintainer wants the one-word form, it is one word.

Two cards filed from this round's findings

objectui#9700 — scripts/check-handler-key-read-sites.mjs exits 0 on a live undeclared read reached through a registration wrapper (from objectui#9447's implementer; re-measured by this seat before filing).
objectui#9701 — a bare-string trigger on collapsible is accepted by both published faces and then fails into the error boundary, because asChild cannot take one. ⭐ It is this card's shape pointing the other way — objectui#8236 was declared inert, actually live; that one is declared acceptable, structurally unrenderable — and it sits on the very key objectui#7767 widened. The declaration half was re-measured by this seat (base.ts:515 names string in the SchemaNode union; disclosure.zod.ts:72 types trigger against it; collapsible.tsx:24 is asChild); the render half is attributed and ⛔ not re-run here, because the shared checkout has no node_modules and that zero was not read.

The implementer's two other notes are correctly dispositioned and this seat adds nothing to them.

⛔ NOT clearable by this seat — the fuse fires

Clause-②: yes (step 2 narrows a published accept set), so a contract review is owed before this PR can be readied. references/contract-review.md:50 binds every seat that issues a clause ② verdict to read its own served tier first, and ⛔ a seat below tier does not self-review. This seat serves claude-opus-5; CONTRACT_REVIEW_TIER is claude-fable-5-1. ⇒ below tier, so no in-seat review.

Route, per :53-54: an isolated at-tier review subagent, fed only the card, the existing rulings and the PR — ⛔ never this seat's conclusions, ⛔ never the dispatch order, ⛔ never the implementer's report. Its tier is taken from its own transcript's per-request model stamp; ⛔ get_session measures the dispatching session and is not cross-evidence. The verdict is adopted verbatim or voided entirely, and needs:contract-review stays on this card until then — ⛔ queueing outside is the safe state.

⛔ Enqueue is HELD, twice over

The review above has not run.
CI on head c48c4a101: 36 returned = 36 total_count ⇒ complete; 28 success / 3 skipped / 0 failures, and 5 still running (Test (shard 1/4)–(4/4), Type Check). Zero failures is ⛔ not a green reading while five required jobs are in flight.

⚠️ mergeable_state: behind — main moved under the round. Correctly left alone: the queue rebuilds on current main, a path this seat confirmed today when PR objectui#9697 enqueued and merged from a base that had moved. ⛔ Not a blocker and ⛔ not a reason to push a sync commit.

Generated by Claude Code

Claim: PM loop round 1
Session: session_01UanLVj6xvbS6puBCewLr8L
Branch: claude/issue-8236-collapsible-open-intercept-retire
Worktree: objectui-issue-8236
Domain: domain:spec
Seat: domain:spec#1
File surface: packages/components/src/renderers/disclosure/, packages/types/src/disclosure.ts, packages/types/src/zod/disclosure.zod.ts, packages/components/src/__tests__/, packages/types/src/__tests__/, .changeset/ (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: see below — dispatch-gates.mjs --tier refuses objectui paths by design (it derives gate families from the tree it runs in), so the at-tier reading for this card is the contract reviewer's own transcript stamp, 85/85 claude-fable-5-1, recorded on PR objectui#9699 comment 5717059216
Clause-②: yes
Thread-read: 5716855005
Serial constraints cleared: none — all 12 open PRs in this repository were read for the file surface above and none intersects it; the merge queue was empty at claim time; no triage-named in-flight sibling card has behaviour this card's pins assert

⛔ This comment REPLACES a malformed one, and the reason is worth more than the fix.

This seat's original claim (5715996881) opened with a markdown heading — ## Claim: … — and carried no Clause-②: line at line-start. check-clause2-carriers --pair 9699 refused the pair at exit 4 and named it exactly: 「A heading-style claim is not a claim comment to this predicate, however complete the reasoning under it」, and 「the enqueue gate's content limb … has nothing to read」.

⭐ The script also names the cause, and it is this seat's: 「Every one of the five misses measured in the filing shift was a line composed from memory」. That is what happened here — the claim above is now copied from the literal template in .claude/skills/pm-dispatch/SKILL.md 〈模板与表〉, ⛔ not composed. The original comment's content (the premise re-verification, the file-face scan) stands and is not withdrawn; what was wrong was its shape, and shape is the whole of a machine-readable predicate.

⚠️ The same run found the second half of the same failure: needs:contract-review is a DUAL carrier (maintainer 2026-08-22, 「两边都挂好」) and this seat hung it on the card only. It is being hung on PR objectui#9699 in the same act as this comment. ⭐ The second carrier is not ceremony — it is the only machine-readable evidence that a gate was cleared rather than stripped: a legitimate clear leaves two removals seconds apart, a strip leaves one, and since 「闸门被剥不是红灯是放行」, without the pair 「被剥」 and 「从未挂过」 are indistinguishable.

Generated by Claude Code

ENQUEUED — clause ② cleared at tier, both carriers cleared in one stroke

2026-09-17T15:40Z · PM seat domain:spec @ objectui · Seat: domain:spec#1 · session session_01UanLVj6xvbS6puBCewLr8L

PR objectui#9699, head c48c4a101850d41e9e847757e3d62e1e2e41da27 — unchanged since the review, so the record on that head is not void.

The bar, in order

gate	reading
governed surface	`Governed Surface Queue Guard` success; `check-governed-queue-guard --test` exit 0 — NOT GOVERNED
clause ②	PASS at tier. Isolated at-tier review subagent, fed only this card, the ruling `5713647032`, the triage verdict `5699709013` and the PR. Tier read from its own transcript: 85 of 85 `assistant` records stamped `claude-fable-5-1` = `CONTRACT_REVIEW_TIER`, zero at any other model ⇒ no mid-review fallback. Record of record: PR comment `5717059216`
`--pair`	`check-clause2-carriers --pair 9699` → exit 0: declaration readable in the fixed spelling, both carriers agree, a review of record names this head
CI	head re-read from the PR object, then its check-runs: 36 returned = 36 `total_count`, 33 success / 3 skipped / 0 anything else, all 9 required contexts green

Carriers cleared in one stroke: card 2026-09-17T15:38:38Z, PR 2026-09-17T15:38:40Z — two seconds apart, which is the signature of a clear rather than a strip.

Queue ref, positively probed: refs/heads/gh-readonly-queue/main/pr-9699-ac716fff422934b9131ed33f3a85a1cf455eec42, whose base is PR objectui#9698's queue head ⇒ the two are stacked. Lit control: 12,609 total refs. ⛔ Never the auto_merge field.

⛔ Two defects of this seat's own, both caught by the gate rather than by the seat

--pair 9699 first exited 4, and both rows were this seat's:

C2 — the claim comment was unreadable. This seat's original claim opened with a markdown heading and carried no line-start Clause-②:. The script's wording is the lesson: 「A heading-style claim is not a claim comment to this predicate, however complete the reasoning under it」 — and it named the cause, 「Every one of the five misses measured in the filing shift was a line composed from memory」. That is exactly what happened. The replacement claim 5717074975 is copied from the literal template in SKILL.md 〈模板与表〉.
C1 — the dual carrier was half-written. needs:contract-review was on this card only; the PR side was bare. ⭐ The second carrier is not ceremony: it is the only machine-readable evidence that a gate was cleared rather than stripped — a clear leaves two removals seconds apart, a strip leaves one — and since 「闸门被剥不是红灯是放行」, without the pair 「被剥」 and 「从未挂过」 are indistinguishable.

⚠️ The same C2 defect was found on objectui#9447 / PR objectui#9698, which was already in the queue. It was ⛔ not pulled: the substance was unaffected (Clause-②: no, derived independently twice, nothing owed a review) while only the evidence was unreadable, and spending a queue cycle to repair a comment's shape buys nothing. A conforming claim was posted there before the merge, and --pair 9698 now exits 0.

⭐ A third defect, found only because the read-back probe was re-drawn. Both replacement claims were written through an unquoted shell heredoc, which performs command substitution. The escaped template fields survived; two prose spans spelled `no` were executed and replaced with nothing. ⚠️ The first read-back probe counted empty backtick pairs, read 0, and looked clean — wrong by construction, because substitution removes the backticks along with their contents and leaves no empty pair to count. The damage was found only by a two-form probe: the intact spelling read 0, the eaten spelling read 1, against a lit control at 1 and an absent control at 0. Repaired, with the cause recorded on the comment itself.

Follow to MERGED. Landing probe, pre-calibrated: open?: never in packages/types/src/disclosure.ts 0 → 1; open: retirementTombstone( in packages/types/src/zod/disclosure.zod.ts 0 → 1; open: _open in packages/components/src/renderers/disclosure/collapsible.tsx 0 → 1; lit control overlay.zod.ts keeps its 7 open: members; lit control Controlled open state tree-wide 17 → 15. ⛔ Never merged / merged_at.

Generated by Claude Code

✅ LANDED — ruling A executed in both steps, verified on `main` with two probes re-drawn

2026-09-17T16:00Z · PM seat domain:spec @ objectui · Seat: domain:spec#1 · session session_01UanLVj6xvbS6puBCewLr8L
Implemented-by: claude/issue-8236-collapsible-open-intercept-retire · Reviewed-by: session_01UanLVj6xvbS6puBCewLr8L

PR objectui#9699 merged through the queue. origin/main is now ee70287e41412ac645f1847eec22e42d56affed0 — byte-identical to this PR's queue ref. Queue empty. ⛔ Not read from merged / merged_at.

probe	predicted	observed
step 2 `open?: never` in `packages/types/src/disclosure.ts`	0 → 1	1
step 2 `open: retirementTombstone(` in `packages/types/src/zod/disclosure.zod.ts`	0 → 1	1
step 1 `open: _open` in `collapsible.tsx`, comments stripped	0 → 1	1 (line 46 of the stripped file)
LIT CONTROL `defaultOpen` in that file, comments stripped	non-zero	2
LIT CONTROL `overlay.zod.ts` `open:` members — dialog / sheet / popover keep the key	stays 7	7
LIT CONTROL `Controlled open state` tree-wide	17 → 15	17 → 15
ABSENT-TOKEN CONTROL	0	0
both new pins	present	present

⚠️ Two probes in the first draw were not readings and were re-drawn rather than reported. open: _open read 2 unstripped — one of those is a docblock mention, and a declaration is code, not prose, so the judgement is the stripped 1. And the Controlled open state probe was written with a broken awk escape that printed nothing; ⛔ that blank was not read as a number. Both re-taken above, with controls.

⭐ The pairing that carries the verdict: the same query reads 15 where it read 17, the two that vanished are exactly the open describes on the two faces, and the seven open: members on the overlay siblings did not move. ⇒ the narrowing landed and stayed inside the one schema the ruling authorised — a scope fence proven by an instrument that was demonstrably able to show the opposite.

What landed

Ruling A, both steps, order preserved in one commit: the collapsible registration stops forwarding an authored open onto the Radix root (a named exclusion, ⛔ not the global metadata strip list, which would have stripped a live prop from dialog / sheet / popover); and open is retired on both published faces with the member kept and the refusal message byte-verbatim from the ruling. Because both steps ride one commit, no commit ever exists with the declaration retired and the takeover still running — the property the fixed order protects.

Clause ② cleared at tier: an isolated at-tier review subagent returned PASS, its tier read from its own transcript at 85/85 CONTRACT_REVIEW_TIER with zero records at any other model. Record of record: PR comment 5717059216. check-clause2-carriers --pair 9699 exit 0; both carriers cleared in one stroke, two seconds apart.

Platform reading — fifth positive today

The card auto-closed (state: closed / state_reason: completed) on Fixes #8236. Fifth independent positive today, against the inherited ledger line claiming keyword closes do not work in this repository. That line is retired; the correction is on the seat post. pm:dispatched removed and the assignee cleared in one label write, read back clean: priority:p3 · domain:spec remain.

Left behind, with carriers

objectui#9701 — a bare-string trigger on this same component is accepted by both published faces and fails into the error boundary, because asChild cannot take one. This card's shape pointing the other way, on the key objectui#7767 widened.
Noted, not filed (the implementer's, and this seat adds nothing): the Radix root renders trigger='[object Object]' / content='[object Object]' DOM attributes — the node's own slot keys riding the same rest-spread, pre-existing and unrelated to open; and two files describing the spread channel as 「the same channel that made CollapsibleSchema.open live」, which stays true about the channel and about alert-dialog, so nothing was repaired and no sweep was opened.
.changeset/7767's 「The Zod mirror is untouched」 — ruled A by this seat, left alone: key-scoped it stays true, and rewriting another card's declaration on a reading of an ambiguity is scope taken on someone else's document. 承接者:无. The at-tier reviewer reached the same disposition independently.

Generated by Claude Code

fix(collapsible): stop honouring an authored `open`, then retire it o…

finding(spec): CollapsibleSchema.open is LIVE through the rest-spread — it overrides defaultOpen and leaves the trigger frozen (this card's original "read by no renderer" premise is REFUTED) #8236

Description

⛔ Correction — this card's headline premise was false

The mechanism, re-derived

What the card got right: the faces disagree

Authorship census (conclusion unchanged, re-measured)

The decision, re-framed

⚠️ Not measured on this card

Docs

Class — corrected

Activity

os-zhuang commented on Sep 7, 2026

分诊 · 准入通过 (c) 类

准入判定

为什么是 needs:contract-review 而不是 pm:queue

p3 的理由

hotlong commented on Sep 7, 2026

os-zhuang commented on Sep 7, 2026

答异议(pm:retriage)—— 状态给 pm:queue,预挂的 needs:contract-review 同笔摘掉

为什么这不是决策卡

承接者拿到的边界

26 remaining items

os-sales commented on Sep 17, 2026

Claim: domain:spec @ objectui — dispatching the ruled letter A

Premise re-verified on origin/main bbe57fdd52bf69f993decba179a64dfc89b26e97

File face, declared

os-sales commented on Sep 17, 2026

os-sales commented on Sep 17, 2026

ACCEPT on the work — ⛔ but the card is NOT clearable by this seat: clause ② routes it to an isolated at-tier review

Verified by this seat independently, ⛔ not relayed

The one question handed back — ruled A, leave the entry alone

Two cards filed from this round's findings

⛔ NOT clearable by this seat — the fuse fires

⛔ Enqueue is HELD, twice over

os-sales commented on Sep 17, 2026

os-sales commented on Sep 17, 2026

ENQUEUED — clause ② cleared at tier, both carriers cleared in one stroke

The bar, in order

⛔ Two defects of this seat's own, both caught by the gate rather than by the seat

Next

os-sales commented on Sep 17, 2026

✅ LANDED — ruling A executed in both steps, verified on main with two probes re-drawn

What landed

Platform reading — fifth positive today

Left behind, with carriers

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding(spec): `CollapsibleSchema.open` is LIVE through the rest-spread — it overrides `defaultOpen` and leaves the trigger frozen (this card's original "read by no renderer" premise is REFUTED) #8236

为什么是 `needs:contract-review` 而不是 `pm:queue`

答异议(`pm:retriage`)—— 状态给 `pm:queue`,预挂的 `needs:contract-review` 同笔摘掉

Claim: `domain:spec` @ objectui — dispatching the ruled letter A

Premise re-verified on `origin/main` `bbe57fdd52bf69f993decba179a64dfc89b26e97`

✅ LANDED — ruling A executed in both steps, verified on `main` with two probes re-drawn