feat(spec)!: object-gantt markers, object-timeline mapping and both forms' fields take the shape each block reads; five objectui-held contracts reported as forks (#21464, S-objectui-held)

[objectstack-fleet]

Part of #21464
Clause-②: yes (narrowing)

What this does

The S-objectui-held stage of the ComponentPropsMap z.unknown() close-out, per the claim 5976829576, the stage-5 landing record 5976140848, the seat answers 5963787404 and 5969669440, the stage-5 ACCEPT 5975923386 and domain:devx's pointer 5974144504. The stage took its own value census first (below), then wrote down and typed the contracts that have one viable shape. Under the stop valve, it holds the five contracts (seven ledger lines) that have more than one shape, records them in the enumeration pin, and reports them on the card as forks. Read points are at the .objectui-sha pin ab1879721595. Every cited reader is byte-identical at objectui main 94985a92ba; the two cited @object-ui/types declaration files changed elsewhere, and their cited lines are identical.

member	was	now	read point at ab1879721595
object-gantt markers	z.array(z.unknown())	strict { date, label?, color? } entries, date a string (module-private factory objectGanttMarker()), with aliases title / text / name to label and colour / stroke to color	ObjectGantt.tsx:2505 hands schema.markers to GanttView, which re-bases date (GanttView.tsx:913-924), drops a marker whose date does not parse or falls outside the range (:2394-2413), and draws label and color (:2407, :4083-4100, SVG :3320-3332). objectui declares the authored marker as exactly this shape (types/src/objectql.ts:3787-3794, zod mirror types/src/zod/objectql.zod.ts:2738-2746)
object-timeline mapping	z.unknown()	strict { title?, date?, description?, variant? }, each a field name (module-private ObjectTimelineMappingSchema), with aliases from the flat titleField / dateField / startDateField / descriptionField / variantField	ObjectTimeline.tsx:551 (title), :576 (date), :578 (description), :579 (variant). objectui declares the same four strings on the component prop (:254-259) and in TimelineMappingSchema (:144-149)
object-form fields	z.array(z.unknown()), held at stage 3	field-name strings (formFieldNameList()). A { name } or { field } object entry is refused with what to write instead	ObjectForm.tsx:961-981, flatFields.ts:68-79. objectui declares ObjectFormSchema.fields: string[]. The form still draws a STORED { name } entry by its name (:972, flatFields.ts:72) and skips any other object entry with a warning (:978)
object-master-detail-form fields	z.array(z.unknown()), held at stage 3	the same list (one schema, pinned)	handed to the parent form verbatim, MasterDetailForm.tsx:1693

The fields hold's exit. objectstack-ai/objectui#11550's triage ruling (5969880008 there) retired the { name } entry from objectui's authoring faces, and its closing paragraph says this card "types object-form and master-detail fields as names once its .objectui-sha covers this landing". The pin covers both code landings: objectui 806f327 (the fixtures respelled) and dbd1081 (the registration descriptions and the warning text). merge-base --is-ancestor exits 0 for each. The guide's own respelling (0a53c67) is on objectui main only. Measured at both trees, the only refused values are fixtures probing the read (census below).

No new export. The three new shapes are module-private. No new member carries a default or a transform, so each parsed value is the authored one. The gantt marker is a factory the row calls, as masterDetailDetailEntry() is, and not a lazySchema proxy. A bare proxy used as an array element is never forced by alias-integrity.test.ts's walk, which skips def values that are functions. The first full-suite run caught exactly that: alias integrity — coverage listed ui/component.zod.ts:6611 (this object-gantt marker) as unreached. The factory form is green.

Held as forks (the stop valve)

Each of these has two or more viable spec shapes that no existing ruling or seat answer decides. Each keeps its hold. The enumeration pin's ledger now files each under a new fork stage that names the shapes, and §2 checks that every fork line names at least two different shapes. The emptied objectui-held stage leaves STAGES. Each fork is reported in this stage's os-dev-report open_questions with its census.

• object-metric drillDown.report: direction A's premise is disproved by measurement, so it is reported as a fork and takes no shape, as the ACCEPT 5975923386 directs. Premise (1) holds: the one drawn report the census finds (objectMetricDrillDownMembers-8071.test.tsx:282, a dataset-bound summary report with name and label) parses through ReportSchema. Premise (2), that isDatasetBoundReport's two arms are exactly what ReportSchema admits, fails in both directions (measured by parse, this branch's source):

  value	drawer (isDatasetBoundReport)	ReportSchema
  { name, label, type: 'joined', blocks: [{ name }] }, no block binds a dataset	lists the records	ACCEPT
  { dataset: 'sales' }, no name / label / values	draws the report	REFUSE
  { name, label, dataset: 'sales' }, no values	draws the report	REFUSE (custom at dataset)
  { name, label, type: 'joined', dataset, blocks: [...] }	draws the report	REFUSE (custom at dataset)

  The first row is the silent-fallback class this card closes. It comes from ReportSchema itself: a joined block's dataset is optional, although the schema's own refinement comment says each block is "dataset-bound". objectstack validate passes such a report with exit 0, while a bound block naming an undeclared dataset is refused (chart-dataset-unknown). See Acceptance notes.

• object-form customFields: objectui's runtime FormField (types/src/form.ts:1770) is open (an index signature beside forty-five members, :1906), and eight of its members are the grid widget's snake_case keys (min_rows, allow_add, …), which this package's camelCase rule for config keys does not admit as written.

• object-form and object-master-detail-form sections: objectui#11550 KEPT the inline runtime field in a section ("shape 3", sectionFields.ts:369-370, declared ObjectFormSection.fields: (string | FormField)[]), so this waits on customFields's fork for its third entry arm.

• object-timeline items: a feed entry's content is child schema nodes (a slot position the page walks would judge, or an opaque member), and the arm an entry must match is chosen by the parent's variant (objectui's row-level refinement, or a plain union of the arms).

• action:group / action:menu members: measured from the reads, the key set is mostly action:button's, keyed by type. It also takes keys the rows leave undecided: outcomeMessages (recorded on action:button as "a contract decision, not a pin re-measure"), a member className, the member's own properties.params bag (static-params.ts:142-160) and endpoint (refused on the rows since finding(spec): action:button / action:icon accept endpoint, the key ActionSchema refuses with the prescription endpoint → target: one concept, two verdicts in one release #21005).

The object-kanban conditionalFormatting hold (held-for-decision) is not in this stage's member list and is untouched. Its carrier, objectstack-ai/objectui#11522, is now in state completed (2026-10-03), so its exit may be readable. That is for the seat.

The census

A writer is a value written on the block: a page-component node (an object literal naming the type, flat or in properties, a literal annotated or asserted with the block's objectui type, a direct parse through the row), the block's React component with the member as a prop or inside schema={{…}}, or the argument of a same-file helper that mounts one (helpers found through a block literal or the block's JSX in their body, including one spread from a const; positional parameters resolved at every call site). Values resolve through same-file constants and spreads. Instrument: a TypeScript-AST walk over .ts .tsx .js .jsx .mjs .cjs .mts .json .yaml .md .mdx (fenced code parsed). For the typed members, a text search for the member key in every file naming the block found what the walk does not reach, and each hit was read by hand. Cross-check: the walk reproduces stage 5's drillDown population exactly (26, 25 static).

corpus	markers	mapping	fields, both forms
objectstack 7d0781482d (examples/, packages/, content/, skills/, apps/, docs/, 9306 files)	0	1 · parses	4, all master-detail · all parse
objectui pin ab1879721595 (whole tree, 10267 files)	9 · 8 parse · 1 refused	9 · all parse	73 · 55 parse · 7 refused · 11 not static
objectui main 94985a92ba (7564 files; the release deleted 2726 changesets)	same 9	same 9	73 · 56 parse · 6 refused · 11 not static
hotcrm 4054ec2680, cloud b2d7a7f6f8	0	0	0

objectstack writers: packages/spec/src/ui/component-element-navigation-17987.test.ts (mapping: { title, variant }). For master-detail fields: the showcase's project-workspace.page.ts, content/docs/protocol/objectui/layout-dsl.mdx, and two test copies (packages/lint/src/validate-component-props.test.ts, packages/spec/src/ui/component.test.ts). All are field names.

objectui refused values, under the writer test (5966636964):

refused value	where	drawn or probe
markers: [{ date: 5 }]	types/src/__tests__/gantt-declared-keys.test.ts:168	compile-time refusal probe (@ts-expect-error)
fields: [{ field: 'note' }], [{ field: 'sent_at' }]	plugin-form/src/__tests__/objectFormFieldsMembers-8071.test.tsx:140, :199	probes: the form skips the entry with a named warning
fields: [{ name: 'note' }] ×2	objectFormFieldsMembers-8071.test.tsx:177, :182	probes: objectui's own mirror refuses it; a STORED one still draws
master-detail fields: [{ field: 'note' }]	topLevelFieldsWarnCoverage-8847.test.tsx:110	probe: the warning fires
master-detail fields: [{ name: 'note' }, 'status']	topLevelFieldsWarnCoverage-8847.test.tsx:258	probe: a STORED entry still draws
fields: [{ name, label, type, required }, …] (pin only)	skills/objectui/guides/page-builder.md	the retired guide example, respelled to names on objectui main

Not writers: five markers arrays mount GanttView, the runtime chart, directly (GanttView.virtual.test.tsx ×2, GanttView.dateOnlyZone-10866.test.tsx, the plugin README's GanttView example, demo/main.tsx). Fourteen fields matches are object definitions or permission maps whose own fields key the walk read as the block's (ObjectForm.effectiveOps ×3, ObjectForm.managedEdit ×3, ObjectForm.mobileFullscreen ×3, drawerFormSectionDescription-9834, objectFormNumericStep-9574 ×2, LineItemsPanel.fieldWriteGate-10163 ×2). The 11 that are not static are run-time hand-offs (AppContent, useActionModal, RecordFormPage, ViewPreview, StudioDesignSurface, DrawerForm, EmbeddableForm, ModalForm, ObjectForm.tsx:387, MasterDetailForm.tsx:1693, ObjectView). None passes through the component-props gate.

The forks' census, both objectui trees identical. objectstack, hotcrm and cloud author none of these, except one items and three sections (strings) in objectstack.

• drillDown.report: 3 values. The one drawn report parses through ReportSchema; the two it.each probes ({ note }, the retired objectName form) are not drawn and are refused.
• customFields: 27 values (24 static, 31 entries), every entry keyed name with label and type. One carries group and one defaultValue, keys outside FormField's forty-five that ride its index signature. No entry writes a snake_case grid key.
• sections: object-form 165 values (99 static); 192 string entries, 3 { field } entries and 7 shape-3 inline runtime fields (plugin-form/README.md's wizard, submitTargetRefusal.test.tsx ×3). Master-detail: 13 values, strings only.
• items: 14 values (11 static): 8 feed entries and 5 gantt rows, none with content.
• action:group / action:menu: 40 / 19 values (12 / 6 static, 15 / 6 members). The keys used are name, label, type, locations, target, bodyExtra, bodyShape, objectName, and one autoTrigger in a host auto-trigger test. None uses an undecided key.

Changes

• packages/spec/src/ui/component.zod.ts: objectGanttMarker() (with the reason it is a factory), ObjectTimelineMappingSchema, formFieldNameRefusal and formFieldNameList() with their docblocks. The four members are typed. The held members' docblocks now record each fork, re-read at ab1879721595: the drill report with the measured disagreement, customFields, both sections, items and actionMemberList. The gantt and timeline row docblocks no longer say markers / mapping stay open.
• packages/spec/src/ui/component-objectui-held-typed-members.pin.test.ts (new): §1 15 byte-identical parses of the census writers, plus the absent-member case; §2 17 refusals by code and path, plus the two prescriptions, the no-prescription control and the alias pointers; §3 each shape's exact member set, and the two forms' fields answering identically, messages included; §4 the three D3 ids.
• packages/spec/src/ui/component-props-unknown-members.pin.test.ts: four lines leave the ledger (markers[], mapping, both fields[]). Seven become fork lines naming their shapes (customFields, both forms' sections, items, both action containers' members, and the drill report, which was objectui-held). The objectui-held stage is replaced by fork, and a §2 case is added.
• packages/spec/src/ui/component-form-family-typed-members.pin.test.ts: the header records this stage's outcome.
• packages/spec/src/migrations/entries/semantic/18.ui-object-gantt-markers-typed.ts, 18.ui-object-timeline-mapping-typed.ts, 18.ui-object-form-fields-names-typed.ts (new), and packages/spec/src/migrations/registry.ts: three step-18 rationale fragments at orders 74, 75 and 76 (73 is the last taken on main), inserted where each id sorts. The semantic region is regenerated by gen:migration-registry. No D2 conversion and no RETIRED_KEYS_BY_MAJOR row: page-component properties is not parsed on the save or load path, and the refused values are nested member values.
• Regenerated by check:generated --fix, which proved only these two stale: content/docs/references/ui/component.mdx and docs/audits/2026-07-unknown-key-strictness-ledger.counts/ui.md (ui/ 196 → 198 sites, +2 strict).
• .changeset/21464-component-props-objectui-held-typed.md: @objectstack/spec minor, BREAKING banner, the Clause-② line, FROM → TO, the census, and the ADR-0087 marker registering the three ids.

Measurements

Head 723df54741, base 7d0781482d. Heavy runs went through scripts/pm/os-verify-lock.sh, and every exit code was captured before any pipe.

• Red first, on the published @objectstack/spec@17.6.0 (the npm tarball, ComponentPropsMap[type].safeParse). Each of these is ACCEPTED there: markers: [{ date: 5 }], [{ label: 'Deadline' }], [{ date, title }]; mapping: 'subject', { titleField: 'code' }, { title: 5 }; object-form fields: [{ name, label, required }], [{ field: 'note' }], [5]; master-detail fields: [{ name: 'note' }, 'status']. Each is refused on this branch, pinned in §2.
• pnpm --filter @objectstack/spec test at 723df54741: exit 0, Test Files 611 passed (611), Tests 18157 passed, 1 todo, nothing skipped. An earlier run at a88b9f3e4a read 610 passed and 1 skipped: root-entry-type-nameability.pin.test.ts skips by design while dist is stale, and it ran here after a rebuild.
• pnpm --filter @objectstack/spec typecheck at a88b9f3e4a (only a test comment differs at the head): exit 0, check:test-typecheck: OK (52 files / 246 errors / 135 signatures held). tsc -p tsconfig.test.json --listFilesOnly names the three touched pins.
• Build and generated artifacts: the spec build is green, including the eager gen:schema pass. check:generated proved check:docs and check:strictness-ledger stale, --fix regenerated those two only, and the re-check is green. A later turbo run build --filter=!@objectstack/docs built 72 tasks, all successful.
• Consumers: pnpm --filter @objectstack/lint test passed 119 files and 5622 tests. The 5 skipped tests read lint's own built dist, which this worktree had not built yet. After the turbo build, lazy-deps, runtime-lazy-deps and validate-component-props passed 61 of 61. pnpm --filter @objectstack/example-showcase validate printed "Validation passed".
• The public door, both ways (validateComponentProps from lint src over the built spec, one block per page). These report nothing: a marker with date / label / color, a { title, variant } mapping, form and master-detail name lists, and an arbitrary drill report (still held). These are reported: markers.0.date: 5 as component-props-invalid; a marker title as component-props-unknown-key at properties.markers.0.title, its message naming label; mapping: 'subject' as component-props-invalid; mapping.titleField as component-props-unknown-key; a form { name: 'email', … } entry as component-props-invalid at properties.fields.0, with "write 'email', not an object"; and a master-detail { field: 'note' } entry with the section-vocabulary prescription.
• Ablation, one leg per narrowing, on the committed tree (a88b9f3e4a). Each leg used node scripts/ablation-replace.mjs in wrap mode, inside a driver with its own EXIT INT TERM restore trap on the absolute path, an empty hash read as failure, and HEAD blob 1a6b65102a:
  • markers → z.array(z.unknown()): anchor x1 → x0, blob → 1f6bdbd1a9. Red: 9 failed, 86 passed. The enumeration pin's §1 received exactly [ "object-gantt markers[]" ] and its census-equals-ledger control read 88 against 87. The companion pin failed its five marker refusals, the alias pointer and the §3 member set.
  • mapping → z.unknown(): blob → 2ee8a60e25. Red: 8 failed, 87 passed. §1 received [ "object-timeline mapping" ], 88 against 87, plus the four mapping refusals, the alias pointer and §3.
  • object-form fields → z.array(z.unknown()): blob → 5211b43913. Red: 9 failed, 86 passed. §1 received [ "object-form fields[]" ], 88 against 87, plus the four refusals, both prescriptions and the §3 two-forms identity.
  • Restore after every leg: blob equals HEAD and git diff HEAD is empty. The pins import ./component.zod from source, so no build or dist preflight sits between mutation and run.
• Derived gates at 723df54741: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (no paths) derived 114 commands, all exit 0. --ran printed "114 derived, 114 run, 0 NOT-MEASURED, 0 UNRUN". The first pass reddened only check-spec-docblock-symbol-anchors (and its --self-test) on a line anchor in the new pin's comment; it now cites the file. Every command was re-run on the final head.
• check-widening-tells over the branch diff: --declaration no exits 4 with seven T1 tells, every one a key inside a former z.unknown() bag (the marker's three, the mapping's four). --declaration yes exits 0.
• Narrowed lint: eslint --no-inline-config --format json over the 8 changed TypeScript files reports 8 files, 0 errors, 0 warnings. Population comes from eslint's own --print-config: those 8 resolve a config, and the changeset, component.mdx and the counts page resolve undefined. Invariance: eslint.config.mjs never enables type-aware linting (its own text, about line 327), so this diff cannot move a verdict on an untouched file. The full pnpm lint is CI's.
• NOT MEASURED: the Console Pin Gate, the Dogfood Regression Gate and the full pnpm lint, because they are CI-owned; objectui was read at the pin and at main, not built against this spec. CI on this PR was not waited on.

Acceptance notes

• ReportSchema admits a joined report with no dataset-bound block. objectstack validate exits 0 on reports: [{ name, label, type: 'joined', blocks: [{ name, label }, { name, label }] }]. The same report with a bound block naming an undeclared dataset is refused by chart-dataset-unknown. objectui's report renderer sends it past isDatasetReport (plugin-report/src/DatasetReportRenderer.tsx:208-214) to the pre-9.0 presentation bridge (ReportRenderer.tsx:108), which issues no query. The schema's own refinement comment says a joined report "carries its data on blocks (each block dataset-bound)". This is reported in the dev report for the seat to file, and it is the cheapest route out of the report fork.
• mapping.title / mapping.date are second spellings of timeline.titleField / timeline.startDateField, read between those and the flat fallbacks. They are typed here as objectui declares them; whether to retire them is a different question. Noted, not filed.
• A marker date is any string, as objectui declares it and as the same row's holidays / minDate / maxDate are. An unparseable date is still accepted and draws no line. Noted, not filed.
• objectui's next @objectstack/spec bump: its block schemas take these rows by reference (propsBag over stripImportedDefaults(...)), but no objectui test parses a refused value through them. The refused values are fixtures that mount the component or parse objectui's own mirror. objectui source reads none of the four members' narrowed types.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

11 anchor(s) derived from 1 changed package(s); no hand-written page names any of them, so this run has nothing to list — not a clean bill of health. This check sees only pages that NAME a derived anchor: one that documents this change in prose, or enumerates it in an authoring dialect, names none and stays invisible to it on every run.

What this run could not see

• 6 name(s) were too generic to anchor anything (single lowercase words)
• the SDK route bridge reached 54 of 206 client-bound route-ledger rows — the other 152 have no registrar path: tail to select them, so pages documenting THEIR client methods cannot appear above, on this or any run. Of those 152: 0 are remediable by widening that discovery convention (an in-repo file declares the path; the convention did not scan it); 55 are structural — on a ledger where NOT ONE row is declared in-repo, so no discovery change reaches them at any price; 97 are undecided (no in-repo declaration, on a ledger that has other in-repo registrars — absence and an unreadable spelling are not distinguishable here). The rows themselves: node scripts/docs-audit/affected-docs.mjs --bridge-coverage
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 138 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 83e2feeb46be3c22b7139b01114612e93afcf412 → packageMentionDocs.

Which tree this was computed on

This run read content/docs from 0b902788dd7ccc2935c5468004c5db734a782c64 — the merge of head 723df5474159ed4b2100cd3742a6874f7e22fd69 into base 83e2feeb46be3c22b7139b01114612e93afcf412, which is what actions/checkout gives a pull_request run. Not the PR head.

A worktree cut from an older main holds a different content/docs, so re-deriving there can legitimately return a different list — that is a different tree, not a wrong row. To answer on the same tree:

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin 0b902788dd7ccc2935c5468004c5db734a782c64 && git checkout 0b902788dd7ccc2935c5468004c5db734a782c64
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin 83e2feeb46be3c22b7139b01114612e93afcf412 723df5474159ed4b2100cd3742a6874f7e22fd69 && git checkout -B drift-repro 83e2feeb46be3c22b7139b01114612e93afcf412 && git merge --no-ff 723df5474159ed4b2100cd3742a6874f7e22fd69

node scripts/docs-audit/affected-docs.mjs --json 83e2feeb46be3c22b7139b01114612e93afcf412

⚠️ That checkout carried uncommitted changes, so the commit above does not fully identify what was read.

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 37185775211 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

失败的 job（日志抽取，best effort）：

• Console Pin Gate — 失败步骤: Build the Console SPA at the pinned objectui SHA

  ✗ Built console still carries the PUBLISHED @objectstack/spec.
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️ 断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• ⚠️ 本次没有可用的聚合签名（日志里没有能解析出测试文件名的 FAIL 行）—— 这不是「没有同签名的其他 PR」，是这一轮没测到。跨 PR 聚合本次不可用，请手工比对其他 PR 的同类评论。
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 1 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)