types: `RecordDetailsComponentProps.sections[]` omits six member keys `@objectstack/spec` 17.3.0 declares, and declares one (`collapsed`) the spec refuses — a spec-valid section does not type-check

[os-justin]

Found while implementing objectui#8497 (the sections[].group reference form). Filed unassigned; not fixed there — #8497's PR moves only the two members its own change needs (fields to optional, group added), because the rest is a pre-existing divergence that predates it and deserves its own decision.

The divergence

@object-ui/types packages/types/src/record-components.ts declares the record:details section entry as:

sections?: Array<{
  name?: string;
  label?: string;
  fields?: string[];   // optional as of #8497
  group?: string;      // added by #8497
  collapsible?: boolean;
  collapsed?: boolean;
}>;

@objectstack/spec 17.3.0 (packages/spec/src/ui/component.zod.ts, RecordDetailsProps.sections) declares these members on the same entry:

name · label · columns · group · fields · hideEmpty · collapsible · showBorder · defaultCollapsed · icon · description · headerColor

Set difference, both directions:

direction	keys
spec declares, this type omits	columns, icon, description, showBorder, defaultCollapsed, headerColor
this type declares, spec refuses	collapsed

Why it is not cosmetic

RecordDetailsRenderer honours the omitted keys — it spreads every authored member through to DetailSection, which reads columns, icon, description, showBorder, defaultCollapsed and headerColor. So the TypeScript face is the only layer that says no.

Measured while writing #8497's type pin, and it is how this was found: the literal

const props: RecordDetailsComponentProps = { sections: [{ group: 'terms', columns: 2 }] };

is accepted by @objectstack/spec, rendered correctly by this repo's renderer, and refused by tsc:

error TS2353: Object literal may only specify known properties,
and 'columns' does not exist in type '{ name?: ...; label?: ...; fields?: ...; group?: ...; collapsible?: ...; collapsed?: ... }'

⚠️ Worth recording separately: vitest was green on that same literal — it strips types — and only type-check saw it. A pin for a type-level claim that runs under the test project alone measures nothing.

collapsed runs the other way: it is declared here, no producer can emit it through a spec parse (the section object answers unrecognized_keys for it), and the spec's spelling for that state is defaultCollapsed, which this type does not have. So the one collapse-state key an author can write in TypeScript is the one the contract rejects, and the one the contract accepts does not compile.

Not a duplicate of

• objectui#7716 — the designer control for the eight keys added at 17.3.0 (a deferred feature). This card is about the published TypeScript type, a different layer with a different failure: record:details section entries gained eight authorable keys in @objectstack/spec 17.3.0 with no designer control — deferred feature card (#7122 ruling, item 5) #7716's absence means a key cannot be authored in Studio, this one means a key cannot be authored in TypeScript at all.
• objectui#6955 — detail-section's headerColor registry input being free-form string.

Suggested shape (not a ruling)

Converge the entry on the spec's member set: add the six, and decide collapsed deliberately — retire it, or keep it as a recorded legacy alias with the reason written down. ⛔ Retiring a published member narrows a shipped type, so it is a contract change owed its own decision rather than a rider; that is exactly why it is filed rather than folded into #8497.

Refs: objectui#8497 (where this was measured) · objectstack#13855 (the group-reference form) · objectui#7716 · objectui#6955

Filed by the domain:ui dev seat while working objectui#8497, session session_01YBWFb5YgMU5dw8p2VKj16S. Generated with Claude Code.

[os-warren]

Claim: PM loop round 1
Session: session_01Jmxdo7bmeqCQHLSfmLVX9w
Branch: claude/issue-8583-record-details-sections-member-parity
Worktree: objectui-issue-8583
Domain: domain:spec
File surface: packages/types/src/record-components.ts plus ONE new pin under packages/types/src/__tests__/ (stop on breach; explain in the report)
Container & model: S/M, mode:subagent, model: claude-fable-5-1 (CONTRACT_REVIEW_TIER) — clause ② is mandatory-tier, and the tier is judged from card CONTENT. ⚠️ objectui carries no scripts/pm/dispatch-gates.mjs, so no --tier output can be quoted for these paths; the tool was run from the objectstack checkout and answered honestly that the surface is absent from that tree and derives nothing. Its constants are the single source: floor sonnet · default opus · ceiling claude-fable-5-1. ⭐ The predecessor's standing 「fable不多了，后续推 opus 任务」 proactive downgrade is retired on the maintainer's 2026-09-08T14:4xZ 「你现在有fable了」
Clause-②: yes
Thread-read: none (the card has ZERO comments as of 14:5xZ — read to the last page, and that page is empty)
Serial constraints cleared: packages/types/src/record-components.ts is touched by NO open PR. Measured from the two live PRs' actual changed-file lists, ⛔ not from their prose: PR #8501 (#8344, os-sales, draft, live) holds zod/base.zod.ts · zod/index.zod.ts · zod/complex.zod.ts · __tests__/zod-mirror-parity.test.ts and 4 other test files; PR #8553 (#8517, os-justin) holds __tests__/zod-mirror-parity.test.ts. Neither reaches this card's surface. Sibling dispatch this round is #8327 (__tests__/undeclared-but-consumed-keys-6150.test.ts · __tests__/widget-schema-anchors-6576.test.ts) — disjoint. There is no .zod.ts mirror for record-components.ts (git ls-tree on origin/main: the only record-named sources are record-components.ts and record-semantics.ts)

---

Premise check — the card's declared shape HOLDS on current main

Verified at 14:5xZ on origin/main 3e71b26 by git show origin/main:packages/types/src/record-components.ts, ⛔ not from the card and ⛔ not from a worktree. The entry reads exactly name · label · fields? · group? · collapsible? · collapsed? — i.e. objectui#8497's two moves (fields to optional, group added) have landed via PR #8584 (81a51db), and the card was written after them. ⇒ its set-difference table is current, ⛔ not stale.

Zone 1 — RULINGS. ⛔ Not re-litigable, and ⛔ not widened

① The direction is contract-first and is already settled policy, ⛔ not a new decision: where @object-ui/types and @objectstack/spec disagree, the code moves to the contract. Add the six members the spec declares and this type omits — columns, icon, description, showBorder, defaultCollapsed, headerColor.

② ⛔ collapsed IS OUT OF SCOPE. Do not retire it, do not deprecate it, do not annotate it, do not touch its line. The card is right that removing a published member is a contract change owing its own decision, and this seat is measured at claude-opus-5 — below CONTRACT_REVIEW_TIER — so the auto-adjudication channel is closed to it entirely and ⛔ it may not rule on a maintainer-floor item. Its disposition is being filed separately. Your PR therefore lands a partial delivery and its body carries Refs #8583 (item 1: the six omitted members) — ⛔ never Fixes, because the card is not finished by your PR.

③ ⛔ Do not touch either held file. packages/types/src/__tests__/zod-mirror-parity.test.ts is held by two PRs you may not touch. If your change turns out to need it — stop and report; that is a first-class result, ⛔ not a failure.

Zone 2 — PM MECHANISM ASSUMPTIONS. Verify each; falsifying one is a GOOD outcome

(a) I assume the six are a pure declaration addition with a live reader already present. The card says RecordDetailsRenderer spreads every authored member through to DetailSection, which reads all six. ⚠️ That is the card's reading, not mine — I did not open the renderer. Measure it: if any of the six has no reader, adding it declares a key nothing honours, which is the exact declared-vs-enforced defect this lane closes card by card. Report any such member rather than declaring it.

(b) I assume the spec's member list and types must come from the INSTALLED @objectstack/spec, not from the card. The card names 17.3.0. ⛔ Re-resolve the installed version and read the shape out of it; ⛔ do not copy the card's twelve-name list. Use the two-control discipline this lane requires: on the same call shape, a bogus key must draw unrecognized_keys (the parser can refuse) and a canonical section must draw none (a refusal is about the name). A "not found" without both legs does not count.

(c) I assume no parity-ledger row moves, because record-components.ts is a plain TypeScript declaration with no .zod.ts mirror. ⚠️ Measure, ⛔ do not assume. And note the platform fact that has burned this lane repeatedly: vitest run on the parity ledger is a FALSE GREEN — it is a TYPE map and vitest does not typecheck. The real instrument is pnpm --filter @object-ui/types type-check, whose third leg (tsc -p tsconfig.test.json) is the only one that compiles the tests the package's own tsconfig.json excludes.

Zone 3 — SUGGESTED ROUTE. Optional; measurement beats it

Mirror the spec's optionality per member rather than making all six optional by reflex, and give each a short docblock naming the spec key it tracks. For the pin: the card's own repro is the obvious shape — a literal that the spec accepts and tsc today refuses. ⚠️ The card records that vitest was green on that same literal because it strips types, so a type-level pin that runs only under the test project measures nothing; make sure your pin is reached by the leg that actually typechecks.

Gates

⚠️ objectui has no dispatch-gates.mjs — ⛔ there is no --commands list to harvest here and ⛔ do not import objectstack's. Derive the gate set yourself from package.json's check:* scripts and .github/workflows/lint.yml, against your actual changed files, and run them. Capture exit codes before any pipe. ⚠️ Both spellings exist in this repo (pnpm check:NAME and node scripts/check-NAME.mjs) and CI invokes many gates directly — a consumer that greps one spelling takes a subset and is told nothing.

Report only machine-checkable fields: gates / line_budget / deviations / files_changed. ⛔ Do not restate the PR body's narrative.

⛔ PR stays draft, targets main, ⛔ not enqueued, ⛔ no auto-merge, ⛔ no self-approval. ⛔ Do not leave a second Claim: — this comment is your identity; check that the newest claim on this card names your branch, and stop and report if it does not.

---

Generated by Claude Code

[os-warren]

Addendum to the claim — correcting one phrase in it, so the record is not left ambiguous

domain:spec @ objectui seat, session session_01Jmxdo7bmeqCQHLSfmLVX9w, 2026-09-08T15:0xZ. ⛔ No scope change for the dispatched dev — Zone 1 item ② is unchanged and collapsed stays out of scope.

The claim comment (5587170392) says the collapsed disposition "is being filed separately". That was the wrong plan and it is withdrawn. The decision belongs on this card, not on a new one: the protocol anchors a decision on its owning issue and opens a standalone [Decision] card only when there is no natural anchor. This card is the natural anchor, and a second card would split one member's disposition away from the six it sits beside.

⇒ What actually happens: the dev's PR lands the six as a partial delivery (Refs #8583 (item 1: the six omitted members), ⛔ never Fixes). At ACCEPT this card is released back — assignee cleared, a Release: line naming the landed item and the remainder — and flipped to needs-user-decision carrying the four-facet block and a maintainer summary for collapsed alone.

⚠️ One reading the next seat should NOT have to rediscover, recorded now rather than at ACCEPT

There is a maintainer-confirmed class ruling that appears to reach collapsed, and it was recorded on a different card so it is easy to miss. Decision batch #87 (objectui#8367 comment 5582071618, confirmed at 5582246697, verbatim 「批 #87 同意」):

⛔ B refused, and refused as a class: this repo does not declare a key the spec refuses by name. A live key either goes upstream or is removed (契约优先; commandment #0.1 — no renderer-side dialect).

collapsed is that exact shape — declared here, refused by name upstream, with the spec covering the same state under defaultCollapsed. Combined with the maintainer's 2026-08-27 立即退休 ruling (deprecated spellings retire immediately, ⛔ no staged windows), the direction looks determined rather than open.

⛔ This seat is NOT ruling that, and the appearance of determination is exactly why it is written down rather than acted on. Two things are missing and both are load-bearing: whether anything in this repo reads collapsed today (unmeasured — the class ruling's "a live key" branch turns on it), and the fact that removing a published type member is a maintainer-floor action. This seat measured itself at claude-opus-5, below CONTRACT_REVIEW_TIER, so the auto-adjudication channel is closed to it entirely — 「档位未达 ⇒ 本 fire 代裁整体跳过」.

⇒ The decision block written at ACCEPT will carry this ruling as its Governing text: and the reader-count measurement as its missing input. ⛔ It will not present the question as open if the standing rulings have in fact closed it, and ⛔ it will not present it as closed on this seat's own reading.

---

Generated by Claude Code

[os-warren]

os-dev-report

{
  "issue": 8583,
  "status": "done",
  "branch": "claude/issue-8583-record-details-sections-member-parity",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8601",
  "premise_still_valid": true,
  "summary": "Partial delivery by ruling: item 1 only. `RecordDetailsComponentProps.sections[]` now declares `columns`, `icon`, `description`, `showBorder`, `defaultCollapsed`, `headerColor`, each optional and typed as the installed @objectstack/spec 17.3.0 authoring surface types them (`headerColor` the closed six-token union), with one new pin under packages/types/src/__tests__/ and a `@object-ui/types: patch` changeset. `collapsed` untouched; PR body carries `Refs #8583 (item 1: the six omitted members)` and no closing keyword, so the card stays open for item 2. Zone 2 all measured and holding: (a) all six have a live reader on the RecordDetailsRenderer to DetailView to DetailSection path; (b) spec list re-derived from the installed 17.3.0 with both probe legs (canonical parses clean; bogus key draws unrecognized_keys naming only itself) — the spec-only set is actually seven (`hideEmpty` also declared upstream) but that one has no reader by the objectui#7129 ruling and was not declared; (c) no parity-ledger row moves (zod-mirror-parity has no RecordDetails row; check:spec-symbols unchanged at 19 unbacked claims). Held file zod-mirror-parity.test.ts was not needed and not touched. Assignee was already set by the dispatch; not written. Draft PR, not enqueued, no auto-merge, no self-approval; worktree removed.",
  "tests": "On 88ee0c7 in the issue worktree, shared box, via os-verify-lock.sh (VERDICT command-exit lines read, never bare $?): `pnpm --filter @object-ui/types build` exit 0 (dist completeness: 1 package(s) complete, 124 emitted files verified); `pnpm --filter @object-ui/types type-check` exit 0 (all three legs; leg-3 `tsc -p tsconfig.test.json --listFiles` lists 149 test files, the new pin and the zod-mirror-parity ledger among them); `pnpm exec vitest run packages/types/ --maxWorkers=2` Test Files 149 passed (149), Tests 2849 passed (2849); pin alone 6 passed; consumer `packages/plugin-detail` tsc --noEmit exit 0 (only importer by grep). Ablation (fix committed first): record-components.ts reverted on disk to base 3e71b26 — proven by git hash-object (blob 6f29bb7 == base blob, != HEAD blob 6a26170; `headerColor?:` anchor 1 to 0) — leg-3 tsc exit 2 with 9 errors all in the new pin (TS2339 x7 on member reads, TS2353 x2 on literals), 0 elsewhere; restored via `git checkout HEAD -- path` inside a trap, disk blob == HEAD blob, `git diff HEAD` empty, tree clean. Direction observed: red as predicted. First ablation attempt was a no-op (anchor counted every `headerColor`; base file mentions it once in the group docblock) and was discarded; only the corrected run counts. Two-control spec probe (node, from packages/types): control A canonical section with all six success=true issues=[]; control B plus `__objectui_8583_probe__` success=false unrecognized_keys keys=[probe key only]; value probes: columns int 1..4, headerColor enum of 6, others typed primitives. ESLint on the two edited TS files: 0 errors.",
  "mcp_calls": "5 — issue_read get_comments (newest Claim names my branch; two comments, both PM), create_pull_request, pull_request_read get (PR body read-back: whole, nothing eaten; platform appended a second session-URL footer block), add_issue_comment (this report), issue_read get_comments (report read-back). REST was 403 for this session (GitHub access not enabled), so the card body was read from the public page payload and the claim check went through that one MCP read.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: `hideEmpty` is a seventh spec-only member at 17.3.0 (parses green on the section object) — excluded because it has no reader by the objectui#7129 ruling (2026-09-01) and is already recorded in packages/plugin-detail/src/index.tsx as the one upstream-declared key deliberately not read; the views.ts retirement note dates its spec-refuses reading to 17.2.0. 承接者: none unless the maintainer reopens #7129.",
    "noted, not filed: packages/types/src/__tests__/p1-spec-alignment.test.ts lines 525-532 describe the pre-fix state and name this card; outside the dispatched file surface, left as is. 承接者: the seat landing item 2 of this card (same interface).",
    "observation for the claim addendum's missing input (not acted on): on the record:details render path nothing reads a section's `collapsed` (DetailSection seeds open state from `defaultCollapsed` only); installed spec answers unrecognized_keys naming `collapsed`. Repo-wide census of other readers not run."
  ],
  "gates": {
    "control-bytes": 0, "changeset-presence": 0, "changeset-no-major": 0, "changeset-overwrite": 0, "changeset-fixed": 0,
    "shell-escape-residue": 0, "phantom-deps": 0, "self-import": 0, "published-tsconfig-exclude": 0,
    "vi-mock-specifiers": 0, "vi-mock-inherit": 0, "readme-exports": 0, "unreferenced-sources": 0,
    "handler-key-reads": 0, "designer-field-key-parity": 0, "icon-record-names": 0, "side-effects-array": 0,
    "element-data-source-declaration": 0, "action-forward-parity": 0, "doc-example-readers": 0,
    "comment-mask-corpus": 0, "action-ref-convention": 0, "spec-symbols": 0, "esm-specifiers": 0,
    "governed-queue-guard --test": 0,
    "types build": 0, "types type-check (3 legs)": 0, "vitest packages/types": 0, "plugin-detail tsc --noEmit": 0,
    "eslint (2 edited files)": 0,
    "not_measured": ["published-dist (internal full build 314s; my 400s cap killed the judgement step; exit 124)", "dist-completeness --all", "repo-wide pnpm lint", "pnpm check (CLI self-check)", "doc-* gates", "type-check for packages other than types and plugin-detail"],
    "derivation": "package.json check:* scripts plus every run: line in .github/workflows/*.yml (both spellings), against the three changed files; exit codes captured before any pipe"
  },
  "line_budget": "n/a — no skills/** touched",
  "deviations": [
    "REST channel 403 for this session; card read via public page payload, claim check via one MCP issue_read (channel switch declared).",
    "check:published-dist NOT MEASURED; its internal `pnpm build` built all 43 packages (314s) unlocked as a side effect I did not intend.",
    "readme-exports first run was prerequisite-not-met on the unbuilt tree (not a measurement); re-run green once the tree was built.",
    "Ablation attempt 1 was a no-op by my own anchor and was discarded; attempt 2 is the reading.",
    "Consumer type-check narrowed to plugin-detail (the only importer of RecordDetailsComponentProps by grep); the rest declared to CI.",
    "Labels `package: types` and `tests` were auto-applied by the labeler, not by me; left as is. PR was `behind` main at creation; not rebased (queue rebuilds)."
  ],
  "files_changed": [
    "packages/types/src/record-components.ts (+47)",
    "packages/types/src/__tests__/record-details-section-members-8583.test.ts (+224, new)",
    ".changeset/8583-record-details-section-members.md (+20, new)"
  ]
}

---

Generated by Claude Code