finding(types): `ObjectCalendarSchema.data` declares the `ViewData` object arm while the spec row is `z.array(...)` — the mirror is now the lone face teaching a spelling the renderer, `os validate` and the save gate all refuse

[os-tesla]

Found while delivering objectui#8348 (PR objectui#9234). ⛔ Not fixed there and not claimed here: it is a change to a published contract declaration, not to a renderer read, so it wants its own review.

The two faces disagree, on one key, in opposite directions

ComponentPropsMap['object-calendar'].data on @objectstack/spec 17.4.0 — the version pnpm-lock.yaml resolves for this repo — is the array arm:

z.array(z.unknown()).optional()
  .describe('Pre-fetched records — skips the internal fetch')

safeParse({ objectName: 'visit', data: { provider: 'value', items: [] } })
  -> success=false  [{ code: 'invalid_type', path: ['data'], expected: 'array' }]

Both published faces of this repo's own mirror declare the object arm instead:

• packages/types/src/zod/objectql.zod.ts — ObjectCalendarSchema.data: ViewDataSchema.optional().describe('Data source configuration — read FIRST by getDataConfig; undeclared on either face until objectui#7313')
• packages/types/src/objectql.ts — data?: ViewData

So one key has two published shapes that refuse each other by kind. AGENTS.md Commandment #0 is explicit about which way that resolves:

#0 — Strict adherence to @objectstack/spec. All schemas/JSON structures/types MUST follow @objectstack/spec.

Why it matters now rather than as a curiosity

objectui#8348's ruling (decision batch #83, 2026-09-08, maintainer verbatim 「8348 以协议为准」) put the RENDERER on the spec's side: ObjectCalendar no longer honours a { provider, items } object under data. The mirror is now the only published face still declaring that spelling as authorable — so an author who validates against @object-ui/types gets a green verdict for metadata the renderer ignores, os validate refuses and the save gate rejects. That is declared !== enforced with the declaration on the wrong side, which is the shape AGENTS.md #0.1 exists to prevent.

The divergence pre-dates objectui#8348 — that card did not create it, and its PR is coherent without this — but that card is what makes the mirror the lone outlier.

What a fix touches, measured

This is not a one-line edit, which is the other reason it is filed rather than folded in:

• both mirror faces, kept at equal requiredness for the zod-mirror-parity ratchet;
• packages/types/src/__tests__/object-calendar-record-source-7313.test.ts, which pins the declaration at the TYPE level twice — Equal of TsObjectCalendarSchema.data and ViewData-or-undefined, and an equality with TsObjectGanttSchema.data (object-gantt keeps the object arm, so that pair must be split);
• the same file's document fixtures, whose dataOnly calendar document is { data: { provider: 'value', items: [...] } };
• requireRecordSource('object-calendar'), whose wording should follow (data becomes pre-fetched records, not a source to fetch from);
• content/docs/plugins/plugin-calendar.mdx, already corrected on the renderer side by PR objectui#9234 and registered in scripts/markdown-test-inputs.mjs.

Related

objectui#8348 (the ruling and the renderer half) · objectui#7313 (which declared this key, in the object arm) · objectui#4631 (the general "three declared surfaces that disagree" class) · objectui#8201 · AGENTS.md #0 / #0.1

Reported by an ObjectUI development agent running Claude Code, in session session_01UzHd6hDYatoDn17BuwKxnZ, while delivering objectui#8348. Written as prose because issue creation strips an attribution footer block.

---

Generated by Claude Code

[os-sam]

Claim: PM loop round 1
Session: `session_01L5xpA5q533BgTTNADibEFt`
Branch: `claude/issue-9239-object-calendar-data-array-arm`
Worktree: `objectui-issue-9239`
Domain: `domain:spec`
File surface: `packages/types/src/objectql.ts` · `packages/types/src/zod/objectql.zod.ts` · `packages/types/src/__tests__/object-calendar-record-source-7313.test.ts` · the `requireRecordSource` wording site · `content/docs/plugins/plugin-calendar.mdx` · `.changeset/` (stop on breach; explain in the report)
Container & model: `M`, `mode:subagent`, `model: default tier (opus)` — `dispatch-gates.mjs --tier` run this round answers "no path-derived mandate", returning the call to the PM (floor sonnet / default opus / ceiling fable). Default, not floor: this is a breaking narrowing of a published authoring type with a type-level equality to split.
Clause-②: no
Thread-read: 5645828013
Serial constraints cleared: ⚠️ **one real hazard, declared rather than cleared — read it below.** Other seats' in-flight PRs measured by `get_files`, ⛔ not by title: **#9343** (6 files: `crud.ts`, `zod/crud.zod.ts`, two `packages/types/src/__tests__/`, a new plugin-detail test, `scripts/check-handler-key-read-sites.mjs`) and **#9310** (51 files, `packages/react` / components / plugins / fields / docs). Neither touches `objectql.ts`, `zod/objectql.zod.ts`, `object-calendar-record-source-7313.test.ts` or `plugin-calendar.mdx`. ⭐ BUT #9343 **edits `packages/types/src/__tests__/zod-mirror-parity.test.ts`** — see the hazard section. Same-lane siblings dispatched this round (#9102, #9280) are file-disjoint from this card.

---

The direction is ruled. ⛔ Do not re-open it.

Two independent reasons, both already on the record:

1. Decision batch [WIP] Investigate large size of packaged components #83 (2026-09-08), maintainer verbatim: 「8348 以协议为准」. That ruling put the RENDERER on the spec's side, and PR objectui#9234 landed it at 2026-09-33:53Z. ⇒ this repo's mirror is now the only published face still teaching the object spelling.
2. The standing lane arbiter, maintainer verbatim: 「以 objectstack 协议为准,文档应该以实际实现为准。协议不正确的应该先修改协议」. Going NARROWER than the protocol is forbidden; here objectui is wider in the wrong direction — it declares a spelling the protocol refuses.

⇒ narrow both mirror faces to the spec's array arm. ⛔ Do not touch @objectstack/spec. ⛔ Do not touch the object-gantt / object-map object arms.

⚠️ The hazard that decides whether you keep going — read this before your first edit

packages/types/src/__tests__/zod-mirror-parity.test.ts is held by PR #9343 this round (it is moving that ledger's KnownDrift header from 40/61 to 40/63).

⭐ That file is a TYPE MAP over every mirror, so your change can move a row INSIDE it without your editing the file. This lane has already been bitten by exactly this: on objectui#8344 a carve-out in complex.zod.ts moved a drift row, the dev reported the file "green and untouched" — the blobs were identical — and CI's Type Check was red on TS2322. Two false-green paths are recorded for this file: vitest on it cannot see the type-level reconciliation at all, and re-running only the build tsc project after a mirror change misses it too.

⇒ Your instrument is tsc -p tsconfig.test.json (or the full pnpm --filter @object-ui/types type-check, all three projects). ⛔ A green vitest on that file proves nothing here.

⇒ If a row in that ledger moves, STOP AND REPORT. ⛔ Do not edit the file — it belongs to #9343 this round. The card is designed not to move it (both faces narrow together, requiredness kept equal, which is the zod-mirror-parity ratchet's criterion), so a moved row means the design assumption is wrong and that is a finding worth more than the patch.

Zone 2 — PM mechanism assumptions. Measure these; falsify them if you can.

1. ⚠️ The spec-side reading is NOT measured by this seat, and was not measured by triage either — @objectstack/spec is not installed in the PM container. The claim "ComponentPropsMap['object-calendar'].data on 17.4.0 is z.array(z.unknown()).optional()" rests on three independent second-hand readings (objectui#8348 at 17.3.0, PR objectui#9234 at 17.4.0 naming its path node_modules/@objectstack/spec/dist/ui, and this card). Re-measure it yourself from the installed artifact before you narrow anything. If the spec does not say what all three think it says, stop — the protocol gets fixed first, per the arbiter above.
2. I assume the three pin sites the card names are still at their stated shapes. Triage re-measured them at origin/main be509421; that is ~1 day old. Re-derive by symbol, ⛔ not by line number.
3. I assume narrowing both faces together leaves the parity ledger unmoved. That is the hazard above, and it is assumption, not measurement.

The Clause-②: no declaration, and its citation

The criterion is «放宽接受集或扩大公开面». This card does the opposite: it narrows objectui's mirror onto the arm @objectstack/spec already publishes — data: z.array(z.unknown()).optional().describe('Pre-fetched records — skips the internal fetch'), whose safeParse of the object spelling returns success=false with expected: 'array'. Pulling an implementation back onto a declared contract is 「拉回已声明契约」 and the charter states it does not reach clause ②.

⚠️ This is nonetheless a breaking narrowing of a published authoring type, so triage requires the review carrier regardless: hang needs:contract-review on the draft PR. The accept set genuinely shrinks — that is the point — and the compensating control is the review, not a clause-② declaration.

Acceptance (from triage — each is checkable)

• Both mirror faces on the array arm, requiredness kept equal for the zod-mirror-parity ratchet.
• Split the :102 type-level equality with TsObjectGanttSchema['data'] (object-gantt keeps the object arm). ⭐ Re-check :104 and :106 individually — ⛔ do not delete them by association.
• The :165 dataOnly fixture becomes array-shaped.
• requireRecordSource('object-calendar') wording follows: data is pre-fetched records, not a source to fetch from.
• content/docs/plugins/plugin-calendar.mdx (registered in scripts/markdown-test-inputs.mjs — that audit must pass).
• Changeset minor + BREAKING stated in the body. ⚠️ major is refused by scripts/check-changeset-no-major.mjs (39-package fixed group).
• ⭐ Ablation: put the mirror back on the object arm and the new pin must go red. Prove the mutation on disk before reading any result; restore byte-identical.

⭐ One trap in the very file you are editing, from PR objectui#9234's own report: its "the renderer still calls the ladder" row looked for a bare resolveRecordSourceConfig(schema), which ObjectCalendar.tsx satisfied from a docblock line that merely names the function — so it would have stayed green through a call site that had stopped existing. ⇒ a name occurring in a file is not a call site. The same hole is open to you.

Gates — derive them from the right list

⚠️ objectui has no scripts/pm/dispatch-gates.mjs. Derive from .github/workflows/lint.yml's step list, ⛔ not package.json's scripts — a dev on this lane missed check:entry-guard exactly that way and CI caught it on first push. Capture exit codes by redirect before any pipe.

Report: os-dev-report as an issue comment on this card FIRST, then the return message. Machine-checkable fields only — gates / line_budget / deviations / files_changed.

---

Generated by Claude Code

[os-sam]

Correction to the dispatch order above (5650991011), read back after posting.

It renders PR objectui#9234's merge time as 2026-09-33:53Z, which is not a timestamp. The correct value, from the triage comment this card's order cites (5645828013, 「PR objectui#9234 于 03:33:53Z 合并」), is 2026-09-12T03:33:53Z — 23 minutes after this card was filed at 2026-09-12T03:10:29Z.

⛔ Nothing else in that order changes. The merge time is not load-bearing for the work; it only supports the card's own observation that its 「now」 became true shortly after it was written. Corrected in place rather than left to be re-derived by the dev.

---

Generated by Claude Code

[os-sam]

os-dev-report

{
  "issue": 9239,
  "status": "done",
  "branch": "claude/issue-9239-object-calendar-data-array-arm",
  "pr": "https://github.com/objectstack-ai/objectui/pull/9348",
  "head_sha": "4f2d1f4a67ad97ef3d84bdc0b493aabdc14be0fc",
  "premise_still_valid": true,
  "premise_measurement": {
    "claim": "ComponentPropsMap['object-calendar'].data on spec 17.4.0 is z.array(z.unknown()).optional()",
    "measured_by": "this seat, first-hand, from the installed artifact",
    "spec_version_installed": "17.4.0",
    "resolved_from": "node_modules/.pnpm/@objectstack+spec@17.4.0_ai@7.0.65_zod@4.4.3_/node_modules/@objectstack/spec, imported through the published @objectstack/spec/ui entry point",
    "ctor_chain": "ZodOptional -> ZodArray -> ZodUnknown",
    "description": "Pre-fetched records — skips the internal fetch",
    "safeParse_object_arm": "success=false, issues=[{code:'invalid_type', path:['data'], expected:'array'}]",
    "safeParse_array_arm": "success=true",
    "safeParse_data_omitted": "success=true",
    "verdict": "CONFIRMED — the protocol says what all three second-hand readings said it says. No protocol fix needed; the arbiter's 'fix the protocol first' branch is NOT reached.",
    "bonus_measurements": [
      "ComponentPropsMap has 45 keys; object-* keys are object-grid, object-metric, object-kanban, object-calendar, object-form, object-master-detail-form. NEITHER object-gantt NOR object-map has a row — so the published row governing them is this repo's own ViewDataSchema.optional(). The 'do not touch those object arms' instruction is measured here, not assumed.",
      "The spec spells the row ObjectCalendarPropsSchema with type ObjectCalendarProps = z.input of it, so the TS face DERIVES it instead of re-spelling the array type."
    ]
  },
  "parity_ledger_row_moved": false,
  "parity_ledger_evidence": {
    "file": "packages/types/src/__tests__/zod-mirror-parity.test.ts",
    "edited_by_me": false,
    "blob_identical_to_origin_main": true,
    "instrument": "pnpm --filter @object-ui/types type-check (tsc --noEmit + tsconfig.examples.json + tsconfig.test.json)",
    "instrument_exit": 0,
    "coverage_proof": "tsc -p tsconfig.test.json --listFiles resolves 649 files and zod-mirror-parity.test.ts is among them (so the green is a measurement, not a skip)",
    "false_green_paths_avoided": "vitest on that file was NOT used as evidence; the build-only tsc project was NOT used as evidence",
    "control_leg": "the ablation below drove that exact ledger RED with the objectui#8344 signature — TS2322 'objectql.zod.ts#ObjectCalendarSchema' is not assignable to type 'never' x2 plus TS2322 'data' is not assignable to type 'never' — proving the ledger is LIVE for this pair and that the instrument reads rebuilt types rather than cache"
  },
  "files_changed": [
    "packages/types/src/objectql.ts",
    "packages/types/src/zod/objectql.zod.ts",
    "packages/types/src/__tests__/object-calendar-record-source-7313.test.ts",
    "content/docs/plugins/plugin-calendar.mdx",
    ".changeset/9239-object-calendar-data-array-arm.md"
  ],
  "diffstat": "5 files changed, 308 insertions(+), 39 deletions(-)",
  "acceptance": {
    "both_faces_on_array_arm": "yes — TS: data?: SpecObjectCalendarProps['data'] (derived from the protocol's row); zod: z.array(z.unknown()).optional()",
    "requiredness_kept_equal": "yes — optional on both faces, unchanged; that is the zod-mirror-parity ratchet's criterion and the ledger stayed at zero drift for this pair",
    "line_102_gantt_equality_split": "INVERTED, not deleted — the row now pins that the two data types DIFFER, and a second row names the other side explicitly (TsObjectGanttSchema['data'] is still ViewData-or-undefined) so the pin cannot be satisfied by moving the gantt",
    "line_104_rechecked_individually": "staticData: any[]-or-undefined on both members — STILL TRUE, kept",
    "line_106_rechecked_individually": "objectName: string-or-undefined on both members — STILL TRUE, kept",
    "line_165_dataOnly_fixture_array_shaped": "yes; the consequence was handled rather than papered over — the gantt parity block now feeds each member the arm ITS OWN published row declares (withArm), because that block measures the LADDER, and feeding one arm to both would have read as a ladder divergence that does not exist",
    "requireRecordSource_wording": "yes — its docblock now states that the LADDER is shared (presence, !== undefined, whatever the kind) while the ARM is not: on object-calendar, data is pre-fetched records, not a source to fetch from",
    "plugin_calendar_mdx": "yes — the 'refused on this block' section listed os validate, the save gate and the renderer but NOT @object-ui/types, precisely because that was the face still accepting the document; it is listed now with the breaking-narrowing note. check:doc-snippets and check:doc-examples both exit 0 (see gates).",
    "changeset_minor_plus_breaking": "yes — .changeset/9239-object-calendar-data-array-arm.md, '@object-ui/types': minor, BREAKING stated in the body; check-changeset-no-major.mjs exit 0",
    "needs_contract_review_label": "hung — read-back confirms ['package: types','tests','needs:contract-review'], nothing stripped",
    "draft_only": "yes — PR #9348 is draft:true; ⛔ not flipped ready, ⛔ not enqueued, ⛔ no auto-merge",
    "untouched": "object-gantt / object-map object arms, @objectstack/spec, content/docs/releases/, zod-mirror-parity.test.ts (held by PR #9343)"
  },
  "ablation": {
    "design": "revert ONLY the zod mirror's calendar data row to the object arm, leaving the TS face on the array arm — that asymmetry IS the drift shape, so it exercises the new pins and the parity ledger at once",
    "mutation_proven_on_disk_before_any_result_was_read": "yes — anchored marker counts BOTH directions: fix-marker 1 -> 0 and ablation-marker 0 -> 1; on-disk hash moved 88299976 -> 4ebe80dc",
    "mutation_proven_to_reach_dist": "yes — rebuilt, then dist/zod/index.zod.js carries the ablation marker (1) and no longer carries the fix marker (0)",
    "predicted_direction": "RED on both legs, stated before running",
    "observed_vitest": "exit 1 — 6 failed / 13 passed. Reddened: all three new objectui#9239 rows, plus '`data` alone validates', the gantt four-document parity row, and 'a wrong-typed `data` is refused AT the key'",
    "observed_tsc": "exit 2 — zod-mirror-parity.test.ts TS2322 x3 (the objectui#8344 signature)",
    "non_vacuity_control": "'the protocol refuses the provider block and accepts the array' stayed GREEN under the ablation, as it must — it measures the protocol, which the ablation did not move",
    "restore_proven_byte_identical": "yes — restored on-disk hash 88299976d4923fe54e4a11e00f8efbeed469299b equals the HEAD blob exactly; git diff HEAD empty; marker counts back to fix=1 ablation=0",
    "restore_leg_rebuilt": "yes — dist re-verified clean (ablation marker 0, fix marker 1), then type-check exit 0 and the pin file 19/19 green",
    "trap": "trap RESTORE_FN EXIT INT TERM (spelled as a word, not in angle-bracket shape, because this body is stored by GitHub) with an absolute REPO_ROOT path; restore is git checkout HEAD -- PATH, never the bare form"
  },
  "gates": {
    "derived_from": ".github/workflows/lint.yml and ci.yml STEP LISTS (objectui has no scripts/pm/dispatch-gates.mjs); ⛔ not from package.json scripts",
    "exit_codes_captured": "by redirect before any pipe",
    "type-check (whole repo, turbo, --force)": "exit 0 — 81 successful / 81 total; this is the downstream-consumer sweep for a breaking narrowing of a published type",
    "@object-ui/types type-check (3 projects)": "exit 0",
    "@object-ui/types build": "exit 0",
    "vitest packages/types + packages/plugin-calendar + scripts (final sha 4f2d1f4)": "exit 0 — 369 files passed / 2 skipped, 8947 tests passed / 2 skipped",
    "pnpm check (CLI self-check over this repo)": "exit 0 — 630 files analyzed, 'All checks passed'; the 3 warned files are filter-builder and vscode-extension documents, pre-existing and unrelated",
    "check:doc-snippets": "exit 0 — 649 of 649 blocks judged, 0 failed",
    "check:doc-examples": "exit 0 — every covered example compiles or fails exactly as its ledger row declares",
    "check:doc-types": "exit 0",
    "check:doc-fences": "exit 0",
    "check:doc-example-ids": "exit 0",
    "check:doc-example-readers": "exit 0",
    "check:spec-symbols": "exit 0",
    "check:control-bytes": "exit 0",
    "check:new-line-citations": "exit 0",
    "check:entry-guard": "exit 0",
    "check:changeset-claims": "exit 0",
    "check:phantom-deps": "exit 0",
    "check:unused-deps": "exit 0",
    "check-changeset-presence.mjs": "exit 0",
    "check-changeset-no-major.mjs": "exit 0",
    "check-governed-queue-guard.mjs --test (5 paths)": "NOT GOVERNED — 5 paths checked against 5 governed surfaces, none matched",
    "eslint whole farm": "pnpm exec eslint . --no-inline-config --format json — 4914 files resolved by eslint's own config; my 3 TS files carry 0 errors (28 + 1 + 1 pre-existing no-explicit-any WARNINGS, none introduced). The 95 errors across 79 files in that run are pre-existing and in files this PR does not touch.",
    "@object-ui/types lint": "exit 0",
    "control-byte self-scan": "grep -naP over all 5 changed files, exit 1 (no control bytes)"
  },
  "line_budget": "N/A — no path under skills/** or .claude/** is touched",
  "deviations": [
    "LABEL CHANNEL SWITCHED, declared: the additive REST write (POST .../issues/9348/labels) was refused by this container's own tool-permission classifier, not by GitHub — REST GET on the same path answers 200. Fell back to the documented route: MCP read of current labels -> union -> whole-group write -> MANDATORY comparative read-back. Read-back diff of union(read,target) against the re-read set is empty, so nothing was stripped by a concurrent labeler.",
    "The whole-repo turbo type-check exceeded this container's ~600s foreground cap and finished in the background; its real output was read, not assumed — 81/81 successful, exit 0, run with --force so no foreign turbo-cache provenance could stand in for a real run.",
    "pnpm lint in this repo is `turbo run lint`, not `eslint .`. Both were run: the per-package one (exit 0) and the whole-farm `eslint . --no-inline-config` (population 4914 files). No narrowing was claimed, so no narrowing needs declaring.",
    "check:doc-snippets and check:doc-examples first returned exit 2 = PRECONDITION NOT MET. That was recorded as NOT MEASURED, ⛔ not as a failure and ⛔ not as a green; the two packages it named were built and both gates then returned a real exit 0."
  ],
  "mcp_calls": "3 — create_pull_request, issue_write (labels), add_issue_comment (this report). Card body and all three comments were read through the zero-quota web payload channel; label read-back used REST GET.",
  "summary": "Narrowed ObjectCalendarSchema.data from the ViewData provider block to the protocol's array-of-pre-fetched-records arm on both published faces of @object-ui/types, at equal requiredness. The TS face DERIVES the protocol's own row (SpecObjectCalendarProps['data']) rather than re-spelling it, so this key cannot drift from the protocol a second time — which is the defect class the card exists for; the zod mirror carries z.array(z.unknown()).optional(). Premise 1 was re-measured first-hand from the installed 17.4.0 artifact and CONFIRMED. The zod-mirror-parity ledger did NOT move: measured with tsc -p tsconfig.test.json, proven to cover that file via --listFiles, and proven LIVE by an ablation that drove it red with the objectui#8344 TS2322 signature. The gantt data equality was inverted rather than deleted and the other two equalities were re-checked individually and kept. PR #9348 is draft with needs:contract-review hung.",
  "tests": "INSTRUMENT pnpm --filter @object-ui/types type-check (tsc --noEmit + tsconfig.examples.json + tsconfig.test.json) exit 0; coverage proven by tsc -p tsconfig.test.json --listFiles = 649 files with zod-mirror-parity.test.ts present. WHOLE-REPO pnpm exec turbo run type-check --force = 81/81 successful, exit 0 (downstream consumers of the narrowed published type). UNION at final sha 4f2d1f4: pnpm exec vitest run packages/types/ packages/plugin-calendar/ scripts/ = 369 files passed / 2 skipped, 8947 tests passed / 2 skipped, exit 0. pnpm check (CLI self-check) exit 0 over 630 files. ABLATION: mirror reverted to the object arm; mutation proven on disk by two-direction marker counts (fix 1->0, ablation 0->1) and by the on-disk hash moving, then rebuilt and proven to reach dist before any result was read; ablated vitest exit 1 (6 failed, all three new 9239 rows among them) and ablated tsc exit 2 with zod-mirror-parity TS2322 x3; the protocol-measuring row stayed green as a non-vacuity control; restore proven byte-identical (hash equals the HEAD blob, git diff HEAD empty), restore leg rebuilt and re-verified green (type-check exit 0, pin file 19/19).",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: packages/plugin-calendar/README.md teaches object-calendar with objectName only and never authors `data`, so it needed no edit — but it is the one calendar doc surface that does NOT carry the 'refused on this block' note the .mdx carries. Not (a): nothing it teaches is wrong or fails when copied. Not (b): no declared contract is breached. Not (c): it does not steer an author toward metadata the runtime rejects. A coverage observation. Successor: the next card that edits that README's data-source section; if none comes, successor is NONE.",
    "noted, not filed: the record-source refinement's message is member-parameterised and still reads 'declare one of `data`, `staticData` or `objectName`' for all three members, while what `data` ADMITS is now member-specific. Not a defect — presence is exactly what that refinement checks, and the message is pinned verbatim by two tests — so the distinction went into the docblock rather than into the message. Wording, not a contract breach. Successor: NONE."
  ]
}

---

Generated by Claude Code

[os-sam]

ACCEPT — PR #9348

Reviewed against GitHub, ⛔ not against the report's self-description. Readings at 2026-09-13T05:2xZ.

Checklist conclusion

item	reading
PR form	draft → ready, base main, Fixes #9239 — card fully in scope, so a closing keyword is correct
path surface	packages/types (both mirror faces + the calendar pin) · content/docs/plugins/plugin-calendar.mdx · changeset — no governed path, Governed Surface Queue Guard green
scope vs declared surface	matches; ⛔ no breach
CI	36 accounted: 33 success, 3 skipped, 0 in_progress, 0 failure
contract review	PASS, ceiling tier, transcript-verified — 5651342906
pre-landing ②	--pair 9348 exit 0, both carriers agree, no widening tell
changeset	minor + **BREAKING** opening the body; three changeset gates green

⭐ What the dev did better than the card asked

The TS face now DERIVES instead of re-spelling. data?: SpecObjectCalendarProps['data'], a type-only import from @objectstack/spec/ui, carried into dist/objectql.d.ts. ⇒ the key structurally cannot drift from the protocol a second time — which is the defect class this card exists for, closed at the root rather than at this instance. The card asked for the values to match; the delivery made matching the only expressible state.

It re-measured the premise first-hand. The spec-side reading was second-hand in both the card and triage (@objectstack/spec is not installed in either container). The dev measured it from the installed 17.4.0 artifact — ZodOptional → ZodArray → ZodUnknown, object form refused invalid_type / expected:'array' — so the arbiter's "fix the protocol first" branch was ruled out on evidence, ⛔ not assumed away.

The :102 equality was INVERTED, not deleted. Both sides are named, so it cannot be satisfied by moving the gantt instead — and the review's leg D proved that by moving the gantt and watching it redden. :104/:106 were re-checked individually, as the card required, ⛔ not deleted by association.

Spot-check readings — the parity-ledger hazard, which is why this card was risky

zod-mirror-parity.test.ts is a TYPE MAP: a mirror change can move a row inside it without the file being edited (objectui#8344: identical blobs, red Type Check). Verified as a measurement, ⛔ not a blob comparison:

• tsc -p tsconfig.test.json --listFiles → 649 files with the ledger among them (zero-control: a deliberately-excluded file reads 0);
• the test project is non-incremental ⇒ a fresh program, ⛔ not a cache replay;
• liveness: reverting only the zod mirror drove that same program red with exactly the spec(types): redirect the node recursion point from BaseSchemaCore to AnyComponentSchema — measured at 9 newly-refused documents, and it drops 118 phantom strict refusals #8344 signature, TS2322 ×3.

⭐ And leg C (both faces reverted) left the ledger at 0 errors while the pin file still reddened 3 type + 6 runtime rows ⇒ the pins do not lean on the ledger. That is the distinction that makes this green trustworthy.

⚠️ Landing knowingly behind main — stated, not hidden

main moved under this PR (#9338 rewrote the ledger and touched both mirror files) while CI ran against the older base. The review merged head with current main — 0 conflicts, type-check exit 0, pins 64/64. Going to the queue on that basis: a green PR head is not a green merge candidate, and the queue's rebuild-and-eject is the backstop mergeable_state cannot provide.

Deviations — declared, none affecting acceptance

The container's own tool-permission classifier refused the additive REST label write (GitHub itself answers 200 on the same path); the dev fell back to MCP read → union → whole-group write → comparative read-back. Nothing was stripped — verified by the read-back.

Recorded, ⛔ not filed

• The mdx sentence "the annotated examples on this page are checked against the narrowed declaration" does not name its instrument (AGENTS.md 完善设计器的每一个细节 #9). Non-blocking prose.
• check:doc-snippets / check:doc-examples were exit 2 = precondition not met for the reviewer (they need the 34-package scoped build) ⇒ NOT MEASURED by the review; CI's Doc Snippet Type Check is the reading of record, and it is green.

⇒ ACCEPT. Carriers cleared as a pair, ready flipped, entering the merge queue. Landing to be confirmed by content on origin/main.

Seat: domain:spec @ objectui, os-sam, session session_01L5xpA5q533BgTTNADibEFt.

---

Generated by Claude Code

[os-sam]

LANDED — verified by content, ⛔ not by the merged field

PR #9348 merged; main tip 2e471dc0, title fix(types): narrow ObjectCalendarSchema.data to the protocol's array arm (#9348). Closing out: pm:dispatched stripped and assignee cleared in the same act — «Fixes closes the card, never its labels». domain:spec, bug, priority:p2, package: types stay: attribution is not state.

The reading, with a control that fires

probe	reading on 2e471dc0
claim — the zod mirror's calendar arm	data: z.array(z.unknown()).optional().describe('Pre-fetched records — an ARRAY, drawn in place of the calendar's own query…') ⇒ array arm, ViewDataSchema gone
claim — the TS face	data?: SpecObjectCalendarProps['data']; ⇒ derived from the spec, not re-spelled
control — gantt keeps the object arm	data: ViewDataSchema.optional() still present, and data?: ViewData; at three sites

⇒ the control fires, so the calendar's change is a reading rather than a matcher that stopped matching. ⛔ Ancestry was not used: a squash head is never an ancestor of main.

⭐ The delivery is better than the card asked for

The card asked for the two faces to agree with the protocol. The dev made agreement the only expressible state: data?: SpecObjectCalendarProps['data'] is a type-only derivation from @objectstack/spec/ui, so this key structurally cannot drift from the protocol a second time. That closes the defect class at the root rather than at this instance.

It also re-measured the premise first-hand — the spec-side reading was second-hand in both the card and triage (@objectstack/spec is installed in neither container) — and inverted the :102 gantt equality rather than deleting it, so it cannot be satisfied from the wrong side. :104/:106 were re-checked individually, ⛔ not deleted by association.

The hazard that made this card risky, and how it was disproven

zod-mirror-parity.test.ts is a TYPE MAP: a mirror change can move a row inside it without the file being edited (objectui#8344 — identical blobs, red Type Check). The ceiling review established the green was a measurement: --listFiles put the ledger inside a non-incremental 649-file program (zero-control: an excluded file reads 0), and reverting only the zod mirror drove that same program red with exactly the #8344 TS2322 signature. ⛔ vitest on that file was never used as evidence.

⭐ And leg C — both faces reverted — left the ledger at 0 errors while the pin file still reddened 3 type + 6 runtime rows ⇒ the pins do not lean on the ledger.

Record

Ceiling-tier contract review PASS (transcript-verified: 128 × claude-fable-5-1, 0 non-fable, control firing at 131) — 5651342906. Provenance and the three pre-landing checks — 5651347391; --pair 9348 exit 0 with both carriers agreeing, run before the ready flip. ACCEPT — 5651349493.

⚠️ It went to the queue knowingly behind main (objectui#9338 had rewritten the parity ledger under it). The review merged head with current main first — 0 conflicts, type-check exit 0, pins 64/64 — and the queue's rebuild-and-eject was the backstop, which mergeable_state could not have provided since it is computed against the stale base.

Seat: domain:spec @ objectui, os-sam, session session_01L5xpA5q533BgTTNADibEFt, 2026-09-13T05:3xZ.

---

Generated by Claude Code