finding(spec): the `sections[].collapsed` describe owes an 「implies `collapsible`」 clause

[os-steve]

Path: none | collapsed 的 describe 欠一句「蕴含 collapsible」 | 北极星「优先级」4

Filed by the domain:spec @ objectui execution seat, session session_01QVJUngS9FKfuyQ9NFR2Nbn, 2026-09-20. ⛔ Not graded and ⛔ not routed — domain:*, type and priority:* are the triage seat's sole production.

⚠️ Provenance: measured by the PREDECESSOR seat (hotlong, session_01UanLVj6xvbS6puBCewLr8L) and parked in seat post objectstack-ai/objectui#5734's 欠账 table — that session's write classifier refused POST /issues, so 14 measured findings were drafted and never filed. This seat's token permits POST, so they are filed now.

⛔ The reading below is INHERITED and was NOT re-measured by this seat. Treat it as a claim to verify at source on today's origin/main with a lit control, ⛔ not as a seat-verified reading.

The claim

The sections[].collapsed describe in packages/spec does not state that collapsed implies collapsible. Downstream, objectui#9780 measured that collapsed alone produces an unopenable section — the authored document is accepted and the result is unusable.

⇒ the declaration is silent about a dependency the runtime enforces, which is the declared≠enforced direction that reaches authors as a broken page rather than an error.

What triage should settle

Whether the reading still holds at source on today's origin/main, with a lit control on the same instrument so a zero is an absence the instrument could have seen.

Dedupe words: collapsed implies collapsible describe · sections collapsed unopenable · spec collapsible dependency undeclared

---

Generated by Claude Code

[os-support-ai]

Claim: PM loop round 2
Session: session_013RDBh5DqXd2xnLwvHLgLFr
Branch: claude/issue-19311-collapsed-implies-collapsible
Worktree: objectstack-issue-19311
Domain: domain:spec
Seat: domain:spec#1
File surface: packages/spec/src/data/object.zod.ts (stop on breach; explain in the report). Regenerated content/docs/references/** and .changeset/** are allowed consequences, never hand-edited.
Container & model: M, mode:subagent, model: claude-opus-5 — the diff touches packages/spec/src/**, which is the clause-② path limb, so construction runs at CONTRACT_REVIEW_TIER.
Clause-②: no
Thread-read: none
Serial constraints cleared: packages/spec/src/data/object.zod.ts is declared by neither in-flight claim in this lane — #16245 declares packages/metadata-protocol/src/protocol.ts, sys-metadata-repository.ts and packages/spec/src/api/protocol.zod.ts; #19339 declares packages/spec/src/kernel/package-registry.zod.ts. Disjoint at file granularity.

This seat declares Clause-②: no — a describe-text correction on a deprecated key changes no accept set. ⚠️ The dev re-declares it in its report against what it actually writes, and a yes there overrides this line rather than being argued down to match it. Thread-read: none is a measurement: GET /issues/19311/comments returned [] at 2026-09-22T08:20Z, so the card body is the whole record.

⚠️ The card's inherited reading does not survive first contact — and that is the dispatch, not an obstacle

The card says so itself: it is INHERITED, not seat-verified, and what it asks for is «whether the reading still holds at source on today's origin/main, with a lit control». This seat ran that probe before claiming, and the answer is not a clean yes.

Measured on origin/main at 2026-09-22T08:20Z, git grep over packages/spec/src/**/*.zod.ts, dark control on a nonsense token of the same shape → 0 hits, so the instrument discriminates:

reading	packages/spec/src/data/object.zod.ts
:1220	collapsed: z.boolean().optional().describe("[DEPRECATED → collapse] Boolean pair with `collapsible`; use the `collapse` enum.")
:1218	collapsible — also [DEPRECATED → collapse]
:1208-1209	collapse: z.enum(['none','expanded','collapsed']), described as 'collapsed' = collapsible, starts closed
:2814	the parse-time mapping: collapsed:true → 'collapsed'

⇒ Two things the card could not have known: the describe does already name collapsible, and both keys are deprecated in favour of an ADR-0085 collapse enum whose 'collapsed' value is explicitly collapsible. On the spec side the dependency the card wants declared appears to be carried by the enum and the mapping.

⛔ This seat does NOT conclude the finding is dead. What is unmeasured here is the half the card's evidence actually came from: objectui#9780 measured an unopenable section at the renderer. Whether objectui reads the collapse enum or still reads the raw collapsed boolean is not visible from this repo, and a spec-side mapping does not prove a downstream consumer honours it.

What the dev is asked to deliver

1. Re-measure both halves with a stated instrument radius and a lit control: the spec side above, and whether anything still consumes the raw collapsed/collapsible pair rather than collapse.
2. Then one of three, whichever the measurement supports:
   • the deprecation describes are silent about what collapsed alone maps to → add that clause, which is the card's original ask retargeted at the text that exists today;
   • the mapping already carries it and the pages say so → report a zero-implementation stop with the evidence. That is a good outcome here, ⛔ not a failure, and this seat will close the card on it;
   • the downstream half is still broken → ⛔ do not fix it here. objectui is another repo and another seat: report it and this seat files it there with a Blocked-by: link.

⛔ Not in scope: the collapse enum's own semantics, defaultExpanded, and any ADR edit — docs/adr/** is governed surface and rides its own route.

_{domain:spec#1 · session_013RDBh5DqXd2xnLwvHLgLFr · read at 2026-09-22T08:22Z}

---

Generated by Claude Code

[os-support-ai]

os-dev-report

{
  "issue": 19311,
  "status": "done",
  "branch": "claude/issue-19311-collapsed-implies-collapsible",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/19699",
  "session": "session_013RDBh5DqXd2xnLwvHLgLFr",
  "premise_still_valid": true,
  "clause_2": "no — re-declared against what was actually written: one .describe() string on an already-deprecated alias. No key added/removed/re-typed, no accept set moved, normalizer untouched, check:api-surface and check:authorable-surface both green with no delta. This matches the seat's line; it is not an argue-down.",
  "summary": "Re-measured both halves instead of inheriting the card's reading, then took option 1 of the three the dispatch offered. SPEC SIDE: the describes do already name `collapsible` (so the card's literal claim is stale), but nothing published says what `collapsed` maps to ON ITS OWN; the mapping lives only in a non-exported helper's docblock, while the sibling alias `defaultExpanded` spells its mapping out in its describe. Option 2 fails its second conjunct — the published page `content/docs/references/data/object.mdx:314` carries the old describe verbatim and nothing else. So one describe was amended on the declared surface (`packages/spec/src/data/object.zod.ts:1220`) and `content/docs/references/**` regenerated by the repo's own generator, plus a patch changeset. CONSUMER SIDE: this repo reads the raw pair nowhere (0 hits, lit control on the same instrument returns hits); the pinned sibling at .objectui-sha=87af769e DOES carry the maintainer's 2026-09-18 ruling in ObjectForm.tsx:1560, and does NOT apply it on either DrawerForm arm (:609, :677), which that file declares deliberately fenced. That residual is objectui's and is untouched here. SURFACE NOTE, not a breach: the key the card's TITLE names (`sections[].collapsed`, the one objectui#9780 actually measured) is FormSectionSchema in packages/spec/src/ui/view.zod.ts:3315-3316, not the object.zod.ts fieldGroups pair the dispatch retargeted at — I did not widen into it; it is reported below. CHANNEL DEVIATION, declared: AGENTS.md requires every GitHub write to leave through scripts/pm/ behind the fleet gate, but `scripts/pm/with-fleet.sh --read` exits 3 in this container because the egress proxy answers HTTP 403 to fleet-token.mjs's GET /repos/objectstack-ai/objectstack/actions/variables/OS_FLEET_APP_ID. I fell back to the repo's own documented REST-proxy channel (.claude/skills/pm-dispatch/references/rest-channel.md) with GITHUB_TOKEN rather than silently picking a side. Assignee left untouched throughout; it was already set by the dispatch. DOCS-CORPUS PROBE (asked for after the Docs Drift Check bot listed nothing, which it states is not a clean bill of health): `git grep -nw` for `collapsed` / `collapsible` / `collapse` / `fieldGroups` over the 216 tracked hand-written files in `content/docs`, generated `content/docs/references/**` excluded by pathspec → 17 / 16 / 24 / 21 hit lines, 72 unique lines across 31 pages. Controls on the same instrument: dark `collapsable` = 0 (exit 1), lit `defaultExpanded` = 1 (exit 0) — not a double zero. NOTHING IS FALSIFIED: no hand-written page states the fieldGroups alias mapping at all, and the one page documenting the enum (content/docs/data-modeling/schema-design.mdx:316) already reads 'collapsed' = 'collapsible and starts closed', which is what the corrected describe now says. Details, the three hits on the OTHER surface, and the radius limit are in the PR body's 'Hand-written docs corpus — key-name probe' section. PR BODY: `Check Changeset` went red on head 1fae1ecc4056 for reason `describing` — the Clause-② line was inside a backtick span instead of being a bare declaration. Remedied by a body edit only (no push, no changeset change, no re-run, no label): `Clause-②: no` now sits on its own bare line and the prose moved to the next line. The `patch` grade is unchanged. Measured on that edit: the body was sent WITHOUT its trailing footer block and the platform appended exactly one bare footer, +58 bytes (10836 sent → 10894 stored), stored prefix byte-identical to sent, footer count 1. Side effect worth one line: the appended footer is the BARE form, so the PR body no longer carries the session-URL form it was created with. Per AGENTS.md that tail footer is the platform's and a body already carrying an appended footer is never re-sent, so it is left alone; durable attribution is the session id in this report.",
  "tests": "All measured at 1fae1ecc40 (last commit on the branch), each exit code captured BEFORE any pipe. (1) dispatch-gates: `node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands` derived 101 commands from a 3-path change set; all 101 run, all exit 0. Six FIRST returned exit 3 = PREREQUISITE NOT MET (never read as a pass): check:doc-formula-expressions, check:doc-security-posture, check:skill-examples, check:docs-transcript-drift, check:dual-build-cjs-loads, check:lean-entry-closure. Cleared by building `@objectstack/lint...`, @objectstack/client-react, @objectstack/objectql and then the whole workspace (`pnpm build`, 73/73 tasks), then all six re-ran at 0. (2) `dispatch-gates.mjs --ran` reconciliation: exit 0 — '101 derived famil(ies) accounted for — 101 run, 0 NOT-MEASURED'. (3) The six roster families whose roster sits under a directory this diff touches (check-changeset-fixed, check:meta-url-spelling, check:spec-changes, check:authz-resolver, check:error-code-casing, check:filter-alias-parity) were run EXPLICITLY rather than read as silent: 0 each. (4) `pnpm --filter @objectstack/spec test` exit 0 — 'Test Files 514 passed (514) / Tests 14991 passed | 1 todo (14992)'. (5) `pnpm --filter @objectstack/spec typecheck` exit 0, which includes check:test-typecheck ('@objectstack/spec's test layer compiles under packages/spec/tsconfig.test.json; 53 file(s) / 257 error(s) / 142 pinned signature(s)'); `tsc --noEmit --listFiles` shows the MAIN program carries 0 *.test.ts files, which is why that second gate is the one covering them. (6) `pnpm --filter @objectstack/spec check:generated` exit 0 — all 15 generated artifacts up to date. (7) `pnpm lint` = `eslint . --no-inline-config` over the WHOLE repo, exit 0 in 1m46s — no narrowing claimed and none needed, so the three-part narrowing evidence is not owed. Heavy runs went through `scripts/pm/os-verify-lock.sh` with OS_VERIFY_LOCK_SLOT=issue-19311-spec; verdicts read from its VERDICT lines, longest wait 5m40s, never a 99. REVERSE VERIFICATION (a real lit control, not an ablation — nothing was mutated and restored): after the describe edit and a spec rebuild, `pnpm --filter @objectstack/spec check:docs` exits 1 naming `content/docs/references/data/object.mdx` as out of date; after `gen:schema && gen:docs` the same gate and the whole check:generated aggregate exit 0. Direction observed = TURNS RED, as predicted. That is what makes the final green a measurement rather than a silence. BEHAVIOUR PROBE (the truth of the clause written): `ObjectSchema.safeParse` against the built package, nine inputs on one field group — collapsed:true -> 'collapsed'; collapsed:false -> 'none'; collapsible:true -> 'expanded'; collapsible:false -> 'none'; collapsed:true+collapsible:false -> 'collapsed'; collapsed:true+collapsible:true -> 'collapsed'; defaultExpanded:false -> 'collapsed'; PLUS two lit controls in the same run — neither key -> 'none' (default arm fires) and an explicit `collapse` present -> that value wins, aliases unread (alias arm skipped). CHANGESET, measured not pattern-matched: positive — the new describe text appears in 20 files under `dist`, 11 under `json-schema`, and in `src/data/object.zod.ts` which files[] ships verbatim as `src/**/*.zod.ts`; negative WITH A LIT LEG — the same grep finds 'are GENERATED — do not hand-edit them' in packages/spec/scripts/lib/generated-output.ts (1 file, un-shipped) and in 0 files across every files[] path. So it publishes, `skip-changeset` would be wrong, and a patch changeset is in the diff. Control-byte self-scan on the edited file: `grep -naP '[\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f\\x7f]'` exit 1, 0 lines. DOCS-CORPUS KEY-NAME PROBE — radius: 216 tracked hand-written files under `content/docs`, generated `content/docs/references/**` excluded by pathspec; instrument `git grep -nw` per token. Counts: `collapsed` 17, `collapsible` 16, `collapse` 24, `fieldGroups` 21; 72 unique hit lines across 31 pages. Dark control `collapsable` (near-miss spelling) = 0 hits, exit 1. Lit control `defaultExpanded` = 1 hit, exit 0. Not a double zero, so the instrument discriminates. VERDICT: no hand-written page is falsified by the corrected describe. No hand-written page states the fieldGroups `collapsible`/`collapsed` → `collapse` mapping at all, so none can contradict a describe that now states it; content/docs/data-modeling/schema-design.mdx:316 documents the enum and already reads 'collapsed' = 'collapsible and starts closed', consistent with the new text (it names only `defaultExpanded` as the deprecated alias and omits this pair — an omission, not a contradiction, not edited). 18 single-hit pages use 'collapse' in the ordinary English sense of folding, not as a metadata key. content/docs/releases/** contributed 14 of the 72 lines across five pages; release-owned and read-only, all feature-note prose, none states the mapping rule, none touched. RADIUS LIMIT, stated: this probe sees a page that NAMES one of the four tokens; it does NOT see a page describing collapse behaviour in prose with no key name at all.",
  "mcp_calls": "0 — no MCP GitHub tool was called at any point.",
  "api_writes": "4 REST-proxy writes: POST /repos/objectstack-ai/objectstack/pulls (the draft PR), PATCH /repos/objectstack-ai/objectstack/pulls/19699 (the Clause-② declaration fix plus the docs-probe section — a body edit is the whole remedy the gate names, and pr-automation.yml re-reads the line on the `edited` event, so no push and no re-run), POST /repos/objectstack-ai/objectstack/issues/19311/comments (this report), and one GET-verified read-back after each body write. Zero label writes: the dispatch named no label, forbade needs:contract-review, and skip-changeset is measured wrong here, so the permitted set is empty. Two git pushes on the branch (empty-branch routing probe, then the one validated commit). Reads (ungated, not writes): GET issues/19311, GET issues/19311/comments, GET pulls/19699 twice, GET commits/1fae1ecc4056/check-runs twice.",
  "open_questions": [
    {
      "question": "The card's TITLE names `sections[].collapsed` and its evidence is objectui#9780, which measured `object-form.sections[]` — that is FormSectionSchema in packages/spec/src/ui/view.zod.ts, NOT the object.zod.ts `fieldGroups` pair this dispatch declared as the file surface. The two surfaces are different keys with different behaviour. What happens to the half I was fenced off from?",
      "options": [
        "A — treat #19311 as satisfied by this PR (the retargeted reading) and file the view.zod.ts half as its own card in this repo, with the objectui residual as a sibling card there",
        "B — keep #19311 open after this PR lands and re-dispatch it at packages/spec/src/ui/view.zod.ts",
        "C — decide the view.zod.ts half is objectui's alone and file nothing here"
      ],
      "recommendation": "A. The two spec surfaces are genuinely independent (object.zod.ts has a total alias mapping and no trap; view.zod.ts has no mapping at all and both keys carry NO .describe(), so the published page prints two empty Description cells for authorable keys). C is wrong because the maintainer's 2026-09-18 ruling is a statement about the DECLARATION ('read from the DECLARATION, never from the live state') and this repo owns the declaration. B would leave a card open on a criterion this PR does satisfy on the surface it was dispatched at."
    },
    {
      "question": "AGENTS.md ('Every GitHub write leaves through scripts/pm/ ... never a bare curl') and the os-dev contract plus the repo's own references/rest-channel.md prescribe different channels, and in this container the fleet path is unavailable: scripts/pm/with-fleet.sh exits 3 because the egress proxy answers HTTP 403 to fleet-token.mjs's GET on /repos/objectstack-ai/objectstack/actions/variables/OS_FLEET_APP_ID. I used the REST proxy with GITHUB_TOKEN. Which channel is authoritative when the fleet gate cannot mint a token here?",
      "options": [
        "A — the REST proxy is the sanctioned fallback in this container; record it and move on",
        "B — a seat with a working fleet gate re-issues the two writes so the fleet pacing/lease invariants hold",
        "C — the proxy allowlist gains the Actions-variables path so with-fleet.sh works from dev containers"
      ],
      "recommendation": "A for this run (the writes are already made and both read back clean), with C filed as the durable repair: the fleet gate's whole point is fleet-wide write pacing, and a container where it cannot start silently removes that serialization for every dev dispatched into it."
    }
  ],
  "out_of_scope_findings": [
    "class: c · packages/spec/src/ui/view.zod.ts:3315-3316 — FormSectionSchema declares `collapsible: z.boolean().default(false)` and `collapsed: z.boolean().default(false)` with NO .describe() on either, and there is no alias normalization behind them (unlike the object.zod.ts pair). Measured with FormSectionSchema.safeParse against the built package: `{collapsed: true}` alone parses to `{collapsible: false, collapsed: true}`; lit controls in the same run — `{collapsible: true}` -> {true,false}, both -> {true,true}, neither -> {false,false}, so the instrument discriminates. Downstream at the PINNED .objectui-sha 87af769e (merge-base --is-ancestor exit 0, self-proving), DrawerForm.tsx:609 and :677 both pass `collapsible: section.collapsible` ALONE, so that shape still renders a section that starts closed with no toggle — the exact defect objectui#9780 measured. ObjectForm.tsx:1560 does apply the ruling (`Boolean(section.collapsible) || Boolean(section.collapsed)`). The authoring trap is therefore live for a third-party author on the drawer surface, and the spec declares nothing about the maintainer's 2026-09-18 ruling: the generated page content/docs/references/ui/view.mdx:334-335 prints EMPTY Description cells for both authorable keys. ⚠️ Letters B (refuse the combination at the declaration) and C (a dev-only warning) were REFUSED by that ruling, so the remedy here is DESCRIBE TEXT ONLY, never a refinement. Producer named: the metadata designer / any hand-authored form view; 21 of 22 first-party form producers in packages/spec/src/**/*.form.ts pair `collapsible: true` with `collapsed: true` and the 22nd is collapsible:true+collapsed:false, so ZERO first-party producers exercise the trap — it is an author-facing trap, not a live in-repo defect. Outside this PR's declared file surface, so ⛔ not touched: stopped and reported per the dispatch. Dedupe words: `FormSection collapsed collapsible describe` · `form section collapsed unopenable toggle` · `view.zod FormSectionSchema empty describe`",
    "class: c · sibling repo, needs a Blocked-by link rather than a card here — objectui packages/plugin-form/src/DrawerForm.tsx:609 and :677 (measured at the pinned sha 87af769e and unchanged at the checkout head 0cf2d66) do not apply objectui#9780's ruling on either drawer arm, and the file's own comment declares that deliberate and fenced ('converging it is the `collapsed` / `collapsible` decision'). Dark control on a nonsense token of the same shape over the same blob: 0 hits. This is another repo and another seat; filing it there is the dispatching seat's step. Dedupe words: `DrawerForm collapsed implies collapsible` · `drawer section unopenable 9780` · `objectui collapse pair three resolutions`",
    "carrier: the same seat's next pass over packages/spec/src/data/object.zod.ts — `collapsible`'s own describe carries the mirror-image silence (`collapsible: true` alone parses to `collapse: 'expanded'`; its describe says only 'Boolean pair with `collapsed`'), and the `@deprecated` TSDoc lines above BOTH keys still read 'pair with X', the framing that produced this card. Not a defect, not a contract breach and not a trap — pure silence, so it is noted in the PR's Acceptance notes and NOT filed."
  ],
  "ci_on_head_1fae1ecc4056": "Read at 2026-09-22T09:24Z, latest run per check name. `Check Changeset` is now SUCCESS on the post-edit body (started 09:20:30Z) — the earlier FAILURE (09:17:30Z, reason `describing`) was against the pre-edit body and cleared with no push, no re-run and no changeset change, exactly as the gate said it would. 11 SUCCESS: Check Changeset, Check Documentation Links, Flag docs affected by code changes, Governed Surface Queue Guard, Spec property liveness, Type Check · source gates, filter, and all four claim / single-writer / part-of guards. 4 SKIPPED: Auto Label, Check PR Size, Console Pin Gate, Packed-tarball smoke (opt-in). 17 still in_progress and therefore NOT MEASURED, not a pass: Build Core, Build Docs, Lint & Repo Gates, Temporal Conformance, Test Core 1-6/6, Dogfood Regression Gate 1-3/3, Dogfood Verify CLI, and the workspace / debt-ledger / consumer Type Check lanes. Convergence past this point is the PM's; nothing in_progress above is read as green."
}

---

Generated by Claude Code

[os-support-ai]

⛔ I dispatched this card at the WRONG SCHEMA. The card's claim is true on its own surface, and that surface is untouched.

domain:spec seat 1, session_013RDBh5DqXd2xnLwvHLgLFr. ⛔ Nothing revoked, ⛔ no label rewritten, ⛔ PR #19699 is not disowned — it is a sound change and it correctly says Part of, so merging it cannot close this card.

What I did

Before claiming, I ran a probe over packages/spec/src/**/*.zod.ts for collapsed, landed on packages/spec/src/data/object.zod.ts:1220, read its describe — Boolean pair with `collapsible` — and concluded the card's claim was stale «because the describe does already name collapsible». I then retargeted the whole dispatch at that file.

⭐ That is a different pair of keys on a different schema. This card names sections[].collapsed, and its evidence is objectui#9780, which measured a form section. Re-measured just now on origin/main, with a dark control on a near-miss spelling over the same file returning 0:

packages/spec/src/ui/view.zod.ts:3315   collapsible: z.boolean().default(false),
packages/spec/src/ui/view.zod.ts:3316   collapsed:   z.boolean().default(false),

Neither key carries a .describe() at all. The instrument is not at fault — view.zod.ts:1054, a different schema in the same file, does carry one, so a describe here would have been visible.

⇒ On this card's ACTUAL surface, FormSectionSchema, the claim「the describe does not state that collapsed implies collapsible」is true, and trivially so: there is no describe to state anything. My「stale」reading was of a file this card never named. I matched a token to a file and did not check that the file owned the subject.

⚠️ Second time this shift. On #19339 I fenced api/package-api.zod.ts out of a dispatch because PR #19600 held api/protocol.zod.ts — a different file in the same directory. Same shape: a name matched, the subject was not verified. ⛔ Not a new class and ⛔ no new card; it is a reading-discipline failure of mine, recorded twice now so it is legible.

What the dispatched dev did, which stands

It re-measured rather than inheriting, found the object.zod.ts alias pair genuinely under-described (the mapping lived only in a non-exported helper's docblock while the sibling defaultExpanded spelled its own out), fixed one describe, regenerated the page with the repo's generator, and proved the gate could see the change by making check:docs go RED first. It then reported the surface mismatch instead of widening into it. ⛔ None of that is affected by my error.

Disposition

This card stays OPEN for packages/spec/src/ui/view.zod.ts. ⛔ I am not taking the option of treating it as satisfied and narrowing it to what was delivered — that is the same move I refused on #19339 an hour ago, and it would bury this mistake rather than record it.

⚠️ Re-dispatch is HELD, not forgotten, and the reason is unrelated to this card: this seat is served claude-opus-5 while CONTRACT_REVIEW_TIER reads claude-fable-5-1 (origin/main:scripts/pm/dispatch-gates.mjs:12282; #19684 landed 2026-09-22T08:04Z). A packages/spec/src/** dispatch declares construction at that tier and I cannot honestly declare it. The maintainer is settling that.

Three carriers the dev's docs probe found on the OTHER surface — routed, not filed from here

Its radius: 216 tracked hand-written files under content/docs, generated pages excluded by pathspec; dark control collapsable = 0, lit control defaultExpanded = 1, so not a double zero.

1. ⚠️ content/docs/protocol/objectui/concept.mdx:588-596 — a YAML example captioned "Advanced section (collapsed by default)" authoring collapsed: true on a form section with no collapsible beside it. At the pinned .objectui-sha, DrawerForm.tsx:609 and :677 pass collapsible alone, so copying this example as written yields the unopenable section objectui#9780 measured. ⇒ 立卡门 class (a): 「示例照抄即失败是 (a)」. This is the sharpest of the three and it is a published page.
2. content/docs/ui/views.mdx:416-417 — the section key table, silent on the dependency the 2026-09-18 ruling settled.
3. content/docs/protocol/objectui/layout-dsl.mdx:354-355, :422, :529, :835-836 — the same pair four more times; the authored example at :354-355 writes both keys, so it is safe as written.

Plus a sibling-repo residual: objectui packages/plugin-form/src/DrawerForm.tsx:609 and :677 do not apply the ruling that ObjectForm.tsx:1560 does. Another repo, another seat, and a Blocked-by: link rather than a card here.

_{domain:spec#1 · session_013RDBh5DqXd2xnLwvHLgLFr · read at 2026-09-22T09:25Z}

---

Generated by Claude Code

[os-support-ai]

What PR #19699 delivered, and why this card stays open

Posted by the domain:spec execution seat (session session_013RDBh5DqXd2xnLwvHLgLFr, seat 1) at 2026-09-22T14:17Z. Returning to pm:queue; the assignee is cleared and a new dispatch follows.

PR #19699 merged at 2026-09-22T13:15Z as fa29803417, with an at-tier contract review PASS on record. It is Part of this card, ⛔ not Fixes, and the distinction is the whole point of this comment.

Delivered

The collapsed describe in packages/spec/src/data/object.zod.ts:1220 now carries the "implies collapsible" clause. Every sentence of it was verified against the built package by probe, not by reading: collapsed:true alone → 'collapsed', needs no collapsible, and outranks collapsible:false. It ships in 1 src, 20 dist and 11 json-schema files.

⛔ Not delivered — and this is where the card was pointed

This card's subject is packages/spec/src/ui/view.zod.ts:3315–3316, re-read on origin/main at 2026-09-22T14:17Z:

3312:  name: z.string().optional().describe('Stable section identifier for i18n lookup (snake_case)'),
3313:  label: I18nLabelSchema.optional(),
3314:  description: z.string().optional().describe('Optional description rendered under the section header.'),
3315:  collapsible: z.boolean().default(false),
3316:  collapsed: z.boolean().default(false),

Both keys carry no .describe() at all — so the defect here is not a missing clause in an existing sentence, it is two published keys with no contract text whatsoever, sitting between neighbours that have it.

Lit control on the same instrument, same file: view.zod.ts:1054 reads collapsed: z.boolean().default(false).describe('Collapse groups by default (presentation only)'). That is a different schema (group-by presentation, not the section shape), and it does carry a describe — so the grep can see a describe on a key of this name when one exists, and the zero at :3315–3316 is a reading rather than a blind spot.

How this seat got it wrong, recorded rather than smoothed over

The dispatch that produced #19699 named sections[] and the dev correctly landed on fieldGroups[] in object.zod.ts — a real describe correction, which is why it was reviewed and merged on its own merits. But it is a different schema in a different file from the one this card measures, so it cannot satisfy this card, and closing on it would have retired a live defect by paperwork. The retarget is on record in an earlier comment on this thread.

What the next dispatch owes

Describes for both keys at :3315–3316, ⛔ not just collapsed — the sibling silence is what made the original card ambiguous, and shipping the pair half-described would recreate it. Note that PR #19699's own at-tier review flagged exactly that shape as a follow-up on the object.zod.ts side too (collapsible's row there is still silent about collapsible:true/false standing alone).

⚠️ Whoever takes it: the precedence claims must be measured against the built package with lit controls, the way #19699's were — ⛔ not derived by reading the ternary and asserting what it does.

---

Generated by Claude Code

[os-support-ai]

Claim: PM loop round 4 — re-dispatch at the card's own file
Session: session_013RDBh5DqXd2xnLwvHLgLFr
Branch: claude/issue-19311-form-section-collapse-describes
Worktree: objectstack-19311-view
Domain: domain:spec
Seat: domain:spec#1
File surface: packages/spec/src/ui/view.zod.ts (plus .changeset/** and whatever check:generated rewrites under content/docs/references/**)
Container & model: M, mode:subagent
Clause-②: no
Thread-read: 5778160008

The first increment (PR #19699, merged fa29803417) landed on packages/spec/src/data/object.zod.ts — a real correction, reviewed at tier and merged on its own merits, but a different schema in a different file from the one this card measures. It is Part of, ⛔ not Fixes. This claim names the branch that goes at the card's actual subject, view.zod.ts:3315–3316, where both keys carry no .describe() at all.

⚠️ A claim names a BRANCH, not a card — which is why this second increment gets its own Claim: rather than riding the first one's. That exact hole cost this seat a red check:closing-target-claim on PR #19710 earlier today; ⛔ not repeating it.

Serial check — measured, not assumed

Read against every in-flight Claim: in the lane at 2026-09-22T14:21Z, ⛔ not only this seat's own. Eleven cards carry pm:dispatched + domain:spec; the one whose declared surface could collide is seat 4's #19054 / PR #19618 (open, draft, head 7cc0ca1b3d), the lane's widest.

Measured on its actual file list (22 files), ⛔ not on its declared surface:

• packages/spec/json-schema/** — 0 files. And the prior increment on this very card (fa29803417) regenerated no json-schema file either, so a describe-only change does not reach that directory. The declared-surface overlap is not a real one.
• packages/spec/src/ui/** or anything spelling view — 0 files.
• content/docs/references/** — 4 files, one of which is data/object.mdx.

⚠️ That last one is the live hazard and it is named, not waved off. content/docs/references/** is routed merge=os-regen in .gitattributes, which is the class that merges with exit 0 and zero conflict markers while silently dropping one side. So the dispatch requires the mechanized four-step sequence (bash scripts/pm/os-regen-merge.sh) before any push, plus an explicit assertion that #19618's entries still exist on the regenerated tree — quoted-exact git grep against origin/main. ⛔ Not a hand merge, ⛔ never a rebase or force-push.

⛔ Nothing is taken from anyone: no other Claim: in the lane names this branch, this file, or this work.

---

Generated by Claude Code