[objectstack-fleet]

Requested by Tesla · project thread

Part of #25753 (the cleanup leftovers that #25712's closing record 6432294556 moved there: F18, F19, F20, F23, F27 and the 2h list, plus its own item 3, the import tooling). Part of #25677. One PR, because the 2h list edits .claude/** and so the whole diff is Tier S either way.

Clause-②: no
Metadata-protocol: none

维护者速读

• 改了什么: objectui 并入 ui/ 之后剩下的小件清理,一个 PR:
  • 删掉两个已无人调用的 objectui 遗留脚本(F19 workflow-checks.ts 解析器、F20 跨仓 closer 的结果测试),--changed 模式随它唯一的调用方退役,账本本身保留作审计(F18)。
  • field.externalId 在 liveness 账本里由 live 改判 dead(F23):packages/** 与 ui/** 里没有任何读取方,upsert 都按别的键匹配。作者写 externalId: true 时 os lint 会提示,附一个 patch changeset。
  • spec 源码注释里 objectui: packages/… 的渲染器引用改写为 ui/packages/…(F27)。
  • 车队的 playbook(pm-dispatch references、os-dev、checklist-test、dogfood-verification)不再把 objectui 当兄弟仓,settings 去掉 5 条 objectui 写权限(2h 清单)。
  • 删掉只在切换时跑过一次的 scripts/ui-import/ 锁文件拼接工具,pnpm-workspace.yaml 里引用它的注释改为指向保留该说明的提交。
• 为什么改: objectui 已归档,这些文字和脚本描述的是不存在的仓和流程,会误导执行席。
• 风险与回滚: 无运行时改动;唯一面向作者的变化是 externalId: true 的 lint 警告。回滚 = revert 本 PR。
• 你要做的: 无。.claude/** 是 Tier S,由席位在达档复核 PASS 且全绿后经队列落地。F12(ui/.github/labeler.yml)等你在项目线程里选;F14(CI 文档页重写)单独开 PR。

What changes

F20. ui/scripts/check-cross-repo-closer-outcome.mjs is deleted. Its subject was objectui's cross-repo-issue-closer.yml, which left with ui/.github/workflows/, and only objectui's lint.yml ran it. scripts/check-closing-keyword-parity.mjs drops its NON_PARSERS row (the comment above the list now records all three imported rows as gone).

F19. ui/scripts/__tests__/workflow-checks.ts is deleted. Its parser (readWorkflows, triggerBlock, producedCheckNames) had no caller since 2d-3. Its last export, repoRoot, had two importers (check-lockfile-dedupe.test.ts, check-lockfile-integrity.test.ts); each now computes the same path.resolve(dirname(fileURLToPath(import.meta.url)), '../..') itself. Two retirement notes that named the file now say it was deleted, and check-test-shard-results.mjs's header no longer points at it.

F18. The markdown-test-inputs ledger stays, as an audit: markdown-test-inputs.test.ts holds the adjudicated class equal to what the scanner derives, and check-pending-changeset-literals.mjs resolves markdown literals through its deriveCandidates(). Its --changed mode retires. Its only caller was objectui's ci.yml decision step, and the root Test UI job starts on any ui/** change, so no CI step asks that question any more. The header says so; the history sections below it stay.

F23. liveness/field.json re-grades field.externalId from live to dead. The live grade rested on objectui console metadataConverters.ts, which objectui dc14bcb deleted for having zero importers. Re-measured at 1c63231 across packages/** and ui/**, nothing outside packages/spec reads a field's externalId:

• The Studio field inspector (ObjectFieldInspector) edits it, and MetadataService.ts copies it into a field summary. Neither shows it to a person as a preview, so the designer-preview ruling does not apply.
• Upserts match on other keys: a seed dataset's externalId (seed-loader.ts, default name) and an import mapping's upsertKey (import-prepare.ts).

The row carries verifiedAt, evidenceScope: in-repo and an authorHint that names those two keys. The flag's .describe() ("Is external ID for upsert operations") now says it is not read and names the same two keys; the generated reference pages (references/data/field.mdx, object.mdx, system/migration.mdx) are regenerated from it, and the three hand-written field tables (fields.mdx, field-types.mdx, validation-rules.mdx) say the same. @objectstack/lint's pin on the set of warned field keys gains externalId. It is a boolean default(false), so the lint warns only on an authored true. The changeset is a patch on @objectstack/spec, because the ledger ships in the package and the advisory warning is what an author sees.

F27. In ui/page.zod.ts and ui/component.zod.ts, the six renderer citations spelled objectui: packages/… become ui/packages/…. Each cited file exists under ui/, and each named symbol is still in it: useElementApiDataSource and PredicateScopeProvider in SchemaRenderer.tsx, parentDataSource and ParentRecordBinding in record-related-list.tsx, BoundRecordProvider in recordBinding.tsx. The one pin on that spelling (expression-scope-data-root.pin.test.ts:194) moves with it. The other 24 lines of the F27 count are objectui @<sha> measurement records, mostly in migrations/entries/retired-keys/18.*. They say where and when something was measured, so they stay as written, the same rule 2e applied to the liveness ledger.

2h list (#26216's follow-up section):

• .claude/skills/pm-dispatch/references/:
  • triage.md, hotcrm.md and decision-analysis.md name ui/ instead of the objectui repo.
  • director.md drops the objectui pin-chain rule. dispatch-runbook.md drops objectui's release-PR title.
  • filing-gate.md drops the "objectui cards waiting for next.N" line, because ui/ consumes the spec through the workspace now, and renames the consumer follow-up line.
  • platform-readings.md drops objectui's labeler and os-verify-lock.sh readings. rest-channel.md drops objectui's nine required contexts. review-checklist.md drops objectui's skip-changeset exception.
  • seat-post-protocol.md: two board producers, two queries, two boards. The console-pin release precondition goes, because there is no pin any more (console.ts, "Unstamped dist").
  • lanes/director.md drops the pin chain from its duty heading, daily summary and pointer; lanes/triage.md no longer alternates two repos; dispatch-runbook.md and platform-readings.md drop "两仓" (both words meant objectstack and objectui).
  • true-green.md: the objectui section becomes ui/, with pnpm --dir ui test.
• .claude/agents/os-dev.md: the force-push, type-check spelling and dispatch-gates lines no longer name objectui. Its objectui labeler and skip-changeset rows go.
• .claude/skills/checklist-test/SKILL.md: the console build is pnpm console:build from ui/apps/console, and the environment fingerprint has no .objectui-sha.
• .claude/skills/dogfood-verification/SKILL.md: the /_console section describes the in-workspace Console build. The pin-drift bullet is replaced, because pnpm console:build stamps no SHA and os dev has no drift to judge (packages/cli/src/utils/console.ts, "Unstamped dist").
• .claude/settings.json: the five objectui write rows (labels POST, pulls and issues PATCH, ready-for-review, auto-merge) go. The repo is archived and refuses writes. Its read rows stay, because the history is still read there.
• scripts/pm/check-skill-line-ratchet.mjs: eight ceilings fall to the landed counts (each with a one-line note). No line was added anywhere, and no YAML was re-wrapped.

Import tooling (#25753 item 3). scripts/ui-import/ (the README recipe and splice-lockfile.mjs) ran once, at the cutover, and nothing calls it. It is deleted. The one comment that cited its step 6, in pnpm-workspace.yaml's happy-dom override note, now says the recipe retired here and names the commit to read it at (1c63231).

Provenance-only mentions stay as they are: the objectui#N citations, cross-repo-coordination.md's merge note, and the hook comments that cite objectui incidents.

Not in this PR

• F12 (ui/.github/labeler.yml) is waiting on the maintainer's choice, which was asked in the project thread.
• The Studio field form's help text for externalId (field.form.ts, "Mark as external ID for upsert operations") and its generated translations still describe the old meaning, and the designer still offers the toggle. Filed as Studio still offers field externalId as an upsert key after it was graded dead #26674 (a ui/ lane change plus the metadata-forms translation pipeline).
• F14 (ui/content/docs/guide/ci-cd-pipeline.md, 3,243 lines that describe objectui's deleted workflows, read by about a dozen tests) gets its own PR.

Verification

node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derives 142 commands for this diff. All of them ran locally on this branch, and all exit 0 on the head:

• check-plugin-teardown-shape --self-test first exited 3: its fixture commit was outside the shallow clone. After fetching it, exit 0.
• check:liveness first exited 1: the generated liveness/state-counts/field.md shard still counted externalId as live. gen:liveness-counts rewrote that one shard (live 100 → 99, dead 1 → 2), and it is committed. No README Notes cell names the field set.
• check:entry-nameability, check:exported-any, check:dts-closure, check:dual-build-cjs-loads, check:generated and check:skill-examples refused at first (no declarations on disk). After env -u OS_SKIP_DTS pnpm build and a spec rebuild, all exit 0.
• pnpm --dir ui exec vitest run scripts/: 5,055 of 5,059 pass. unit-registry-absence-collision needed @objectstack/client's dist (15/15 once built), and two check-i18n-call-site-keys cases timed out at 15 s while other gates loaded the machine (137/137 alone).

Main moved twice while this was open. #26536 and #26569 edited three of the same skill files (main's root-run console:build wording and its ninth required context kept). #26599 retired conditionalRequired, so the lint pin is now ['externalId'] alone, and the regenerated field shard reads 99 / 0 / 0 / 1 / 6 / 106. After each merge, the touched-surface gates, check:generated (after a spec rebuild) and the lint suite (7,389 tests) pass.

Main moved a third time while the PR sat in the queue, and the queue dropped it for a conflict at 14:25Z. #26693 rewrote the director lane's governed-draft lines, the same lines this PR had edited to drop "本仓". Main's version is taken whole, because it no longer names a second repo. lanes/director.md now reads 86 lines, so its ceiling falls 88 → 86 (6f64f2a7). On the merged tree (03c8cf2b), gen:schema, gen:docs and gen:liveness-counts produce no diff. check:generated (after a spec rebuild), check:liveness, the skill line ratchet with its self-test (164 cases), check-settings-deny-roster, check-closing-keyword-parity and the lint liveness test (109) all pass.

The delta review on 6f64f2a7 (FAIL 6447375373) named one must-fix. #26765 had landed on main and pinned lanes/director.md at 87, on the same ratchet line this PR moves to 86. d8d03efd merges origin/main 83c4bd29 and resolves that row to 86. Its comment now reads 88 → 87 (#26765) → 86 (#25753). fab4534b takes should-fix 1: the triage charter's ratchet note no longer says pickup alternates between two repos. No ceiling moves. On the merged tree, the regen produces no diff, and the ratchet with its self-test, check:generated, check:liveness and the lint liveness test pass.

The first contract review (FAIL on 434b014) named those conflicts and three playbook files the objectui edits had left self-contradictory; both are fixed above. Its should-fixes are in too: the externalId ledger row names every ui/ carrier and the preview search, os-dev.md says plainly that hotcrm has no dispatch-gates.mjs, and the Studio follow-up is #26674.

CI's first run caught one thing the derivation does not list: @objectstack/lint's lint-liveness-properties.test.ts pins the set of field keys the shipped ledger warns on. It now expects the shipped warned set; the package suite passes (7,389 tests).

Also run by hand: the skill line ratchet and its self-test (164 cases), check-settings-deny-roster, check-harness-current, check-closing-keyword-parity with their self-tests, markdown-test-inputs --audit, and the ui vitest files the diff touches (118 tests).

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

This PR changes 1 package(s): @objectstack/spec, touching 3 documentable anchor(s). ⚠️ 2 changed file(s) yielded no anchor (packages/spec/liveness/field.json, packages/spec/liveness/state-counts/field.md), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files.

12 hand-written doc(s) NAME something this change touched and may need an implementation-accuracy re-verification:

• content/docs/concepts/metadata-driven.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/data-modeling/external-datasources.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/data-modeling/field-types.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/data-modeling/validation-rules.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/deployment/troubleshooting.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/deployment/validating-metadata.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/getting-started/quick-reference.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/kernel/contracts/data-engine.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/protocol/backward-compatibility.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/protocol/objectql/types.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/protocol/objectui/concept.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/protocol/objectui/layout-dsl.mdx (via PageComponentSchema (symbol, a top-level const), RecordRelatedListProps (symbol, a top-level const object))

⛔ 6 release-owned page(s) also name something this change touched. These are read-only:

• content/docs/releases/v12.mdx (via RecordRelatedListProps (symbol, a top-level const object))
• content/docs/releases/v14.mdx (via RecordRelatedListProps (symbol, a top-level const object))
• content/docs/releases/v15.mdx (via PageComponentSchema (symbol, a top-level const))
• content/docs/releases/v17/17-0.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/releases/v17/17-1.mdx (via FieldSchema (symbol, a top-level const))
• content/docs/releases/v17/17-7.mdx (via PageComponentSchema (symbol, a top-level const))

content/docs/releases/ is RELEASE-OWNED (AGENTS.md "Documentation Guardrails"): release
notes are written centrally at release time, and a code PR that edits them is the exact PR
that guardrail exists to stop. They are still audited — read-only. If one of them is actually
wrong, file an issue or open a dedicated docs-only PR; do not edit it here.

What this run could not see

• 2 changed file(s) yielded no anchor (packages/spec/liveness/field.json, packages/spec/liveness/state-counts/field.md) — pages documenting those are invisible to this run
• the SDK route bridge reached 55 of 207 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 — 147 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 f7c95cd191fc0bc8464eeb7cb53d9121704eb727 → packageMentionDocs.

Which tree this was computed on

This run read content/docs from 37a79d7e165b86fba7d9eac4324872eb9f6fd0b6 — the merge of head fab4534b1d23fb5a8713b4d1de75722ea2bc6642 into base f7c95cd191fc0bc8464eeb7cb53d9121704eb727, 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 37a79d7e165b86fba7d9eac4324872eb9f6fd0b6 && git checkout 37a79d7e165b86fba7d9eac4324872eb9f6fd0b6
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin f7c95cd191fc0bc8464eeb7cb53d9121704eb727 fab4534b1d23fb5a8713b4d1de75722ea2bc6642 && git checkout -B drift-repro f7c95cd191fc0bc8464eeb7cb53d9121704eb727 && git merge --no-ff fab4534b1d23fb5a8713b4d1de75722ea2bc6642

node scripts/docs-audit/affected-docs.mjs --json f7c95cd191fc0bc8464eeb7cb53d9121704eb727

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

Advisory only, and a precision-first one (#9192): a page is listed because it names a
symbol, wire route or SDK method this diff touched — not because it mentions a changed
package. Each row says which anchor put it there, so a wrong row is reportable rather than
merely annoying. To re-verify, run the docs-accuracy-audit workflow scoped to these files:
node scripts/docs-audit/affected-docs.mjs f7c95cd191fc0bc8464eeb7cb53d9121704eb727 → pass the list as
args.docs, on the commit named under Which tree this was computed on.

[objectstack-fleet]

Test Core (6/8) is red on 434b014. The cli leg in it passed (1,610 tests), and the step still exited 1. The leg that failed is outside the 5,000 log lines the job API returns, so I could not read its failure.

I re-derived the shard's item list locally with select-shard-packages.sh and partition-test-shards.mjs --shard 6/8. The other nine packages are plugin-audit, plugin-security, driver-turso, client, plugin-pinyin-search, connector-mcp, service-job, service-sms and embedder-openai. All nine pass on this head: 44/44 turbo tasks, exit 0. This PR touches none of them; its only package code changes are a lint test pin and a .describe() string in spec.

I'm re-running the failed jobs once. If it is red again, I'll treat it as real and dig in.

---

Generated by Claude Code

[objectstack-fleet]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 434b014e017bccc17af28ed322e1a520d56abd8b
Local-runs: probe — git merge-tree --write-tree origin/main 434b014e (no checkout, no build, no test) to name the files behind the API's mergeable: false / mergeable_state: dirty
Scope: packages/spec/src/data/field.zod.ts, packages/spec/src/ui/component.zod.ts, packages/spec/src/ui/page.zod.ts, .claude/agents/os-dev.md, .claude/settings.json, .claude/skills/checklist-test/SKILL.md, .claude/skills/dogfood-verification/SKILL.md, .claude/skills/pm-dispatch/references/decision-analysis.md, .claude/skills/pm-dispatch/references/dispatch-runbook.md, .claude/skills/pm-dispatch/references/filing-gate.md, .claude/skills/pm-dispatch/references/lanes/director.md, .claude/skills/pm-dispatch/references/lanes/hotcrm.md, .claude/skills/pm-dispatch/references/lanes/triage.md, .claude/skills/pm-dispatch/references/platform-readings.md, .claude/skills/pm-dispatch/references/rest-channel.md, .claude/skills/pm-dispatch/references/review-checklist.md, .claude/skills/pm-dispatch/references/seat-post-protocol.md, .claude/skills/pm-dispatch/references/true-green.md + scripts/pm/check-skill-line-ratchet.mjs, packages/spec/liveness/field.json, packages/spec/liveness/state-counts/field.md, packages/spec/liveness/README.md, packages/lint/src/lint-liveness-properties.ts, packages/lint/src/lint-liveness-properties.test.ts, .changeset/25753-field-external-id-dead.md, .changeset/config.json, packages/spec/package.json, packages/spec/src/data/field.form.ts, packages/spec/src/ui/expression-scope-data-root.pin.test.ts, packages/cli/src/utils/console.ts, packages/drivers/driver-sql/src/builtin-column-collision.ts, scripts/check-closing-keyword-parity.mjs, pnpm-workspace.yaml, package.json, ui/package.json, ui/scripts/markdown-test-inputs.mjs, ui/scripts/tests/check-lockfile-dedupe.test.ts, ui/scripts/tests/check-lockfile-integrity.test.ts, ui/packages/app-shell/src/views/metadata-admin/inspectors/ObjectFieldInspector.tsx, ui/packages/app-shell/src/services/MetadataService.ts, ui/packages/app-shell/src/views/metadata-admin/previews/index.ts, ui/packages/plugin-designer/src/FieldDesigner.tsx, ui/packages/plugin-designer/src/MetadataFieldsPage.tsx, ui/packages/react/src/SchemaRenderer.tsx, ui/packages/react/src/element-data-source/useElementApiDataSource.ts, ui/packages/plugin-detail/src/renderers/record-related-list.tsx, ui/packages/plugin-detail/src/renderers/recordBinding.tsx, ui/packages/app-shell/src/console/blocks/productBlocks.tsx

Inputs read: card #25753 body and all 5 comments; #25712 closing record 6432294556; PR #26598 body, 42-file list, net diff origin/main...434b014e (merge-base c8ed4c9); the 63 check-runs on the head (every one success or skipped by design; Test Core (6/8) reads success now). PR author is objectstack-fleet[bot], so no login note is owed. The PR is a draft.

① Derived judgments

Each accept-set or public-surface change the diff implies, judged on the head's code.

1. packages/spec/src/data/field.zod.ts:2324 — externalId keeps z.boolean().default(false); only .describe() is reworded. Accept set unchanged. RIGHT. The old text ("Is external ID for upsert operations") was a false .describe() on a shipped *.zod.ts (review-checklist.md names that a defect), so the reword is owed, not optional.
2. packages/spec/liveness/field.json:385-391 — field.externalId re-graded live to dead, with verifiedAt: 2026-10-27 (not future-dated), evidenceScope: in-repo, an authorHint. VERDICT RIGHT: at the head, git grep externalId across packages/** and ui/** finds no reader of a field definition's externalId. Every match is either the seed dataset key (packages/metadata-protocol/src/seed-loader.ts, packages/cloud-connection/src/marketplace-install-local-purge.ts:198 parses SeedSchema.shape.externalId), an import mapping's upsertKey, a record-level external-id reference value (ui/packages/fields/src/widgets/LookupField.tsx:171,188,981,1215, ui/packages/components/src/renderers/complex/data-table.tsx:291), a key-classification map that classifies the name and reads no value (packages/drivers/driver-sql/src/builtin-column-collision.ts:151, packages/metadata-core/src/object-schema-fls-references.ts:607), or a carrier that edits or copies the flag (ObjectFieldInspector.tsx:1259-1261, MetadataService.ts:629). The key's declared effect is behaviour (upsert matching), so by the README's own table (packages/spec/liveness/README.md:442-445) a designer input cannot settle it live. No registerMetadataPreview('field', …) exists; the only registered preview is object (ui/packages/app-shell/src/views/metadata-admin/previews/index.ts:45) and no preview file reads externalId. Evidence text is incomplete, see ③ should-fix 1, but the grade stands.
3. packages/lint behaviour: os lint now emits liveness-dead-property on an authored externalId: true. RIGHT, and the "only on true" claim holds: lint-liveness-properties.ts:195-199 (isAuthored: booleans count only when true), so the schema default false is never warned. The lint reads the ledger from @objectstack/spec's shipped liveness/ dir (lint-liveness-properties.ts:97-108), so the behaviour change ships with the spec bump; .changeset/config.json keeps spec and lint in one fixed group.
4. packages/lint/src/lint-liveness-properties.test.ts:1681 pins the warned field set as ['conditionalRequired', 'externalId']. RIGHT ON THIS HEAD, WRONG AGAINST origin/main: feat(spec)!: retirement kits whose conversion retired at or below 16.x age out; the strict unknown-key refusal names the protocol upgrade guide #26599 (dbf7e04, merged after the merge-base) deleted the conditionalRequired row from field.json and the tombstone from field.zod.ts, and main's pin now reads []. git merge-tree reports a content CONFLICT in this file. After the merge the pin must be ['externalId']. Must-fix 1.
5. packages/spec/liveness/state-counts/field.md:15 says 99 / 0 / 0 / 2 / 6 / 107. Same situation: main's shard reads 100 / 0 / 0 / 0 / 6 / 106 since feat(spec)!: retirement kits whose conversion retired at or below 16.x age out; the strict unknown-key refusal names the protocol upgrade guide #26599, git merge-tree reports a CONFLICT, and the merged truth is 99 / 0 / 0 / 1 / 6 / 106. Must-fix 1. (field.json itself auto-merges: the two edits are in different rows.)
6. packages/spec/src/ui/page.zod.ts:486,488,501,754 and component.zod.ts:1759,1762 — six objectui: packages/… citations become ui/packages/…; comments only, no runtime change. RIGHT: all five cited files exist at the head and carry the named symbols (useElementApiDataSource, PredicateScopeProvider in ui/packages/react/src/SchemaRenderer.tsx; parentDataSource, ParentRecordBinding in record-related-list.tsx; BoundRecordProvider in recordBinding.tsx; dataSource in productBlocks.tsx). Zero objectui: packages/ strings remain under packages/spec/src; the one pin (expression-scope-data-root.pin.test.ts:194) moved with it. The .objectui-sha 0707b4ae37 measurement record at page.zod.ts:502-503 stays as dated prose per the accepted 2e ruling.
7. .claude/settings.json — exactly five rows removed (objectui labels POST, pulls PATCH, issues PATCH, ready_for_review POST, auto_merge PUT); every other row byte-identical; the eight objectui GET rows stay. RIGHT: nothing at the head references an objectui write endpoint (git grep -E "objectui/(issues|pulls)" over scripts, .claude, .github, ui/scripts hits only the GET rows and two doc-link fixtures), and check-settings-deny-roster governs the deny list, not these.
8. scripts/pm/check-skill-line-ratchet.mjs — seven ceilings LOWERED to the landed counts: os-dev 404 to 402, platform-readings 460 to 458, rest-channel 82 to 80, review-checklist 77 to 76, filing-gate 68 to 66, director 88 to 87, dogfood-verification 157 to 140. Each new ceiling equals wc -l of the file at the head (402, 458, 80, 76, 66, 87, 140); no ceiling was raised; the seven unchanged-ceiling hit files have unchanged line counts (decision-analysis 50, dispatch-runbook 245, hotcrm 53, triage 8, seat-post-protocol 91, true-green 32, checklist-test 238). Not loosened. RIGHT. Note: main's qa(checklist-test): writes through the pm-dispatch doors, every defect filed in the run, auth/authz fixes dispatched at once #26567 raised checklist-test to 246/246 after the merge-base; the ratchet auto-merges.
9. Deletions ui/scripts/check-cross-repo-closer-outcome.mjs, ui/scripts/__tests__/workflow-checks.ts, scripts/ui-import/README.md, scripts/ui-import/splice-lockfile.mjs. RIGHT: no importer or runner at the head. .github/workflows/lint.yml:3204-3205 runs the ROOT scripts/check-cross-repo-closer-outcome.mjs, a different file that stays; ui/.github/ holds only labeler.yml; the two former repoRoot importers compute it locally (check-lockfile-dedupe.test.ts:123, check-lockfile-integrity.test.ts:36); pnpm-workspace.yaml:711-713 now points at the recipe's last commit.
10. scripts/check-closing-keyword-parity.mjs:206-219 — the NON_PARSERS row for the deleted ui harness goes; NON_PARSERS[0] used by the self-test at :640 is still the root harness. RIGHT.
11. ui/scripts/markdown-test-inputs.mjs — --changed and readStdin removed; markdownTestInputsAmong stays exported (used by check-pending-changeset-literals.test.ts:12); readFileSync is still used at :1242. No caller of --changed exists at the head. RIGHT.
12. .claude/skills/dogfood-verification/SKILL.md:76-86 — replaces the .objectui-sha / build-console.sh / check:console-sha bullets with "the dist is unstamped, os dev cannot judge drift". RIGHT against the code: packages/cli/src/utils/console.ts:478 ("Unstamped dist (every pnpm console:build since release(ui-merge): one release lane for both scopes — @object-ui joins the fixed group, @objectstack/console builds from ui/apps/console, the objectui pin is deleted (#25711) #25745): can't prove drift"), no .objectui-sha in the tree, root package.json:27 has console:build and no check:console-sha.
13. .claude/skills/checklist-test/SKILL.md:135 drops .objectui-sha from the fingerprint. RIGHT (file absent at the head).
14. .claude/agents/os-dev.md:73,125,157,309-310 — four objectui lines dropped or re-rooted. Factually right (hotcrm has neither scripts/pm/dispatch-gates.mjs nor check-half-states.mjs; ui/ packages spell type-check). Wording note in ③ should-fix 3.
15. .claude/skills/pm-dispatch/references/* — the hunks themselves drop objectui-as-sibling wording and change no rule's meaning in: decision-analysis.md:47, dispatch-runbook.md:89, filing-gate.md:42 (the dropped next.N line was a count the triage list no longer has a source for), hotcrm.md:15, platform-readings.md (labeler and os-verify-lock.sh readings), rest-channel.md (objectui's nine contexts), review-checklist.md (objectui skip-changeset), true-green.md:18-24 (pnpm --dir ui test matches ui/package.json's test: vitest run). RIGHT in isolation. But three of these files are left self-contradictory by the same hunks: see ④ and must-fix 2.

② Semver level

.changeset/25753-field-external-id-dead.md: @objectstack/spec: patch. MATCHES what the diff publishes: no key added, removed, narrowed or widened; externalId still parses with default(false); the only author-visible change is an advisory warning on an authored true and reworded descriptions. Precedent for patch on a ledger re-grade: packages/spec/CHANGELOG.md:1550 and :6770. The fixed group carries @objectstack/lint at the same bump, so the lint's new warning ships with it.

Clause-②: no — RIGHT (no accept set, served body or error code moves; the describe() is prose on a shipped .zod.ts). Metadata-protocol: none — RIGHT (no /meta door, no conversion, no retirement kit).

③ Boundary flags

Dev flags in the PR body's "Not in this PR":

• F12 ui/.github/labeler.yml — a maintainer decision, correctly left out. Answered.
• Studio field form help text packages/spec/src/data/field.form.ts:392 ("Mark as external ID for upsert operations") — deferred because it re-runs the metadata-forms translation pipeline. The text now contradicts the same package's .describe() and the ledger. The PR cites no card; none exists among the 300 issues created or updated since 2026-10-26. ESCALATED as should-fix 2: the seat files the follow-up card before landing, per os-dev.md rule 3 ("其余立卡并记明已发布面").
• F14 ui/content/docs/guide/ci-cd-pipeline.md — own PR. Answered.

No open_questions block in the body. The "Verification" section's local runs are the dev's self-report and are not evidence here; the head's check-runs are, and all are green.

Must-fix (decide the verdict):

1. The head does not merge with origin/main and its merged truth differs from what it pins. git merge-tree names two content conflicts: packages/lint/src/lint-liveness-properties.test.ts (head pins ['conditionalRequired', 'externalId'], main now expects [] after feat(spec)!: retirement kits whose conversion retired at or below 16.x age out; the strict unknown-key refusal names the protocol upgrade guide #26599 deleted the conditionalRequired row; merged truth is ['externalId']) and packages/spec/liveness/state-counts/field.md (head 99/0/0/2/6/107, main 100/0/0/0/6/106, merged truth 99/0/0/1/6/106). The API agrees (mergeable: false, mergeable_state: dirty). The test's comment at :1677-1680 also narrates conditionalRequired as still in the set. A merge of main with those two files re-resolved produces a substantive new head that owes a delta review.
2. Governed rule text left self-contradictory by the hunks (card aftercare(ui-merge): the first joint 18.x release from the merged repo — both scopes at one version, @object-ui published, docs site deployed — then close #25556 as superseded and the epic (parent #25677, 6/6) #25753 item 4's own done-when is "pm-dispatch references that still say two repos" updated):
   • seat-post-protocol.md:80-82 now say two producers and two queries, while :83 still says release-ready means "三张板都空", :88 says "三仓各自执行 … 前端仓", and :89-91 keep the "pin 读数 / pin 滞后 ⇒ console bump 单 / 两者都不成立 ⇒ 不 cut" release precondition whose mechanism this PR itself declares gone (no .objectui-sha at the head; console.ts:478). The section cannot be executed as written.
   • lanes/director.md drops the cross-repo pin-chain rule (old :81) but keeps its heading "职责四:…pin 链…" at :72, "pin 滞后" in the daily summary at :74, and "pin 前置见 ../seat-post-protocol.md" at :78 — a half-removed rule.
   • lanes/triage.md:5 makes the scope "objectstack(ui/ 在内)" with no app-repo seat, while :7 still alternates "两仓轮替" — in an 8-line charter.
     Each fix is a same-line reword or a deletion, so the ratchet ceilings only lower.

Should-fix (do not decide the verdict):

1. field.json:388 evidence says "The two ui/ mentions only carry it", but ui/packages/plugin-designer/src/FieldDesigner.tsx:343,363,476,494 and MetadataFieldsPage.tsx:71,251,466 also carry it (a published @object-ui/plugin-designer surface); and the README's mechanical rule (README.md:419-437) asks the row to record the preview enumeration result ("no registerMetadataPreview('field'); the object preview reads no externalId"). Verdict unaffected; the row should say what was searched so the next sweep does not re-derive it.
2. File the follow-up card for packages/spec/src/data/field.form.ts:392 and its generated translations (see ③ above).
3. os-dev.md:157 "hotcrm 两者皆无" — "两者" (both) now dangles because the line names only one script; true in fact (cloud has check-half-states.mjs only, hotcrm has neither), so say so in words.
4. Ambiguous repo counts left in hit files, confirm whether they meant objectui: dispatch-runbook.md:69 "两仓全量对账", lanes/director.md:57 "两仓 is:pr is:open is:draft …", platform-readings.md:295 "两仓 CI 并发组".

④ 设计判断

Per hits file, does the change hold against the cited code and the card's acceptance scenario (card #25753 item 3 "retire one-time import scripts"; item 4 "pm-dispatch references that still say two repos updated"; #25712 residuals F18, F19, F20, F23, F27, 2h list)?

• packages/spec/src/data/field.zod.ts — holds. The new describe() names the two real match keys (SeedSchema.externalId, mapping upsertKey) that seed-loader.ts and import-prepare.ts read.
• packages/spec/src/ui/page.zod.ts, component.zod.ts — hold; every re-rooted path resolves at the head with the named symbol present.
• .claude/settings.json — holds; objectui is archived and refuses writes, the read rows are still used by the history readings.
• .claude/agents/os-dev.md — holds in substance (wording note S3).
• .claude/skills/checklist-test/SKILL.md, dogfood-verification/SKILL.md — hold against console.ts:478, the absent .objectui-sha, and root package.json.
• decision-analysis.md, dispatch-runbook.md, filing-gate.md, hotcrm.md, platform-readings.md, rest-channel.md, review-checklist.md, true-green.md — hold; each hunk drops objectui-as-sibling wording and the surviving rule reads the same.
• lanes/triage.md, lanes/director.md, seat-post-protocol.md — DO NOT hold as edited: the hunks remove the objectui token but leave the count words and the pin precondition that encoded objectui's existence, so each file now contradicts itself (must-fix 2). Against item 4's done-when these three still "say two repos" or three.
• The gate edited with the rules it ratchets (scripts/pm/check-skill-line-ratchet.mjs) — holds: lowering only, each to the landed count.

Implemented-by: claude/project-thread-fiwblf
Reviewed-by: session_01AavokzJ5DndAwitDXvKy4U

VERDICT: FAIL

---

Generated by Claude Code

[objectstack-fleet]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: c260fddcd3e507d6949b812144307e562b6ee9c5
Local-runs: none
Scope: packages/spec/src/data/field.zod.ts, packages/spec/src/ui/component.zod.ts, packages/spec/src/ui/page.zod.ts, .claude/agents/os-dev.md, .claude/settings.json, .claude/skills/checklist-test/SKILL.md, .claude/skills/dogfood-verification/SKILL.md, .claude/skills/pm-dispatch/references/decision-analysis.md, .claude/skills/pm-dispatch/references/dispatch-runbook.md, .claude/skills/pm-dispatch/references/filing-gate.md, .claude/skills/pm-dispatch/references/lanes/director.md, .claude/skills/pm-dispatch/references/lanes/hotcrm.md, .claude/skills/pm-dispatch/references/lanes/triage.md, .claude/skills/pm-dispatch/references/platform-readings.md, .claude/skills/pm-dispatch/references/rest-channel.md, .claude/skills/pm-dispatch/references/review-checklist.md, .claude/skills/pm-dispatch/references/seat-post-protocol.md, .claude/skills/pm-dispatch/references/true-green.md + scripts/pm/check-skill-line-ratchet.mjs, packages/spec/liveness/field.json, packages/spec/liveness/state-counts/field.md, packages/spec/scripts/liveness/check-liveness.mts, packages/spec/scripts/liveness/build-state-counts.mts, packages/lint/src/lint-liveness-properties.ts, packages/lint/src/lint-liveness-properties.test.ts, .changeset/25753-field-external-id-dead.md, packages/spec/package.json, packages/spec/scripts/check-generated.ts, packages/spec/src/data/field.form.ts, packages/spec/src/ui/expression-scope-data-root.pin.test.ts, content/docs/references/data/field.mdx, content/docs/references/data/object.mdx, content/docs/references/system/migration.mdx, content/docs/data-modeling/field-types.mdx, content/docs/data-modeling/fields.mdx, content/docs/data-modeling/validation-rules.mdx, packages/cli/src/utils/console.ts, packages/console/package.json, packages/console/build-dist.sh, package.json, ui/package.json, ui/vitest.config.mts, .github/workflows/lint.yml, .github/workflows/ci.yml, .github/workflows/spec-liveness-check.yml

Inputs read: card #25753 body and all 5 comments (6410938536, 6416455214, 6432304442, 6437605220, 6439149364); #25712 closing record 6432294556; PR #26598 body, 42-file list, API state (mergeable: true, mergeable_state: clean, so no merge probe was owed), net diff origin/main...c260fddc (merge-base fe53769), the four delta commits since the FAIL head (db4593b7 merge, 03efc6cd, 2241a642, c260fddc) each read as a change in its own right; the 67 check-runs on the head: 60 success, 7 skipped by design (Check PR Size, Auto Label, Dogfood Verify CLI, Dogfood Regression Gate, Temporal Conformance, Packed-tarball smoke opt-in, 18.x upgrade harness opt-in), none failed. The earlier FAIL record (PR comment 6441767146, head 434b014) is an input; PR comment 6440374717 is the dispatching seat's own note and is not evidence. PR author is objectstack-fleet[bot], so no login note is owed. The PR is a draft (Tier S, .claude/**), which is the lane's landing affair.

Earlier FAIL items

Must-fix 1 (head did not merge with main; pin and counts disagreed with the merged truth): FIXED.

• mergeable_state is clean at the head; merge commit db4593b7 has parents 434b014 and fe53769 (main).
• packages/lint/src/lint-liveness-properties.test.ts:1717 now reads expect([...authorWarnedProperties('field')]).toEqual(['externalId']); the comment at :1703-1716 narrates conditionalRequired as having LEFT the set on spec: retire tombstones, conversions and migration entries older than one protocol major; a strict unknown-key refusal with an upgrade-guide link takes over (maintainer ruling D2 = A) #25767 (the card behind main's dbf7e04 / feat(spec)!: retirement kits whose conversion retired at or below 16.x age out; the strict unknown-key refusal names the protocol upgrade guide #26599), and externalId as the one row in it. authorWarnedProperties (lint-liveness-properties.ts:421-425) returns the warn-map keys of the shipped ledger, so the pin states the shipped truth.
• Counts, derived from the head's packages/spec/liveness/field.json itself (not from the PR's claim): 72 props rows, live 68, planned 3 (settledWhen, conditionalFormatting, referenceDeletedWhen), dead 1 (externalId), no other status. Main's ledger: 72 rows, live 69, planned 3, dead 0, conditionalRequired absent. The only movement is one row live to dead. The gate resolves each walked schema key by ledger entry, then describe marker, else UNCLASSIFIED (check-liveness.mts:60-61) and folds per status (:881-883), and main's shard reads 100 / 0 / 0 / 0 / 6 / 106, so the merged truth is 99 / 0 / 0 / 1 / 6 / 106, which is exactly what packages/spec/liveness/state-counts/field.md:15 reads at the head. The shard-freshness leg of the same gate (check-liveness.mts:736 countsArtifactErrors) runs in CI (ci.yml:3993, spec-liveness-check.yml:58) and is green on the head.

Must-fix 2 (governed rule text left self-contradictory): FIXED in all three files.

• seat-post-protocol.md:80-88: :80 two producers (main repo to the director seat, cloud to its whole-repo seat), :81 "two producers each scan their own backlog", :82 "two repos, two queries", :83 "release-ready means both boards empty", :88 "the two repos each execute and post readings; no blanket acceptance because of the ui/ front end or a separate deployment". The pin-reading precondition (old :89-91) is gone; the file is 88 lines. The section is now executable as written and consistent with console.ts:478-479 (unstamped dist, no drift to judge).
• lanes/director.md:72 heading "职责四:每日摘要、发版板与总表" (no pin chain), :74 daily summary without "pin 滞后", :80 pointer "鲜度节奏、清板三选一见 ../seat-post-protocol.md" (no "pin 前置"), the cross-repo pin-chain rule deleted; :81-82 (platform-version bump cards on new tag families) stay, they never depended on objectui.
• lanes/triage.md:5 scope "objectstack(ui/ 在内)" and :7 "每 fire 最旧优先;最旧未路由超 4h 的卡本轮先清" — one repo, oldest first, the 4-hour starvation guard kept in single-repo form. No "两仓轮替" remains.

Should-fix items from the earlier record (do not decide the verdict):

• S1 (ledger evidence incomplete): FIXED by 2241a642. field.json:384 now names the Studio inspector, MetadataService.ts, ui/packages/plugin-designer/src/FieldDesigner.tsx and MetadataFieldsPage.tsx as carriers, and records the preview search ("no registerMetadataPreview('field', …) exists, and no registered preview … reads externalId").
• S2 (follow-up card for field.form.ts:392): FIXED. Studio still offers field externalId as an upsert key after it was graded dead #26674 is open (bug, priority:p2, domain:spec, area:studio), names packages/spec/src/data/field.form.ts:392, the generated translations and FieldDesigner.tsx:476; the PR body's "Not in this PR" cites it. field.form.ts:392 at the head still carries "Mark as external ID for upsert operations", as the card says.
• S3 (os-dev.md:157 dangling "两者"): FIXED, now "姊妹仓未必有 scripts/pm/dispatch-gates.mjs:hotcrm 就没有。"
• S4 (three ambiguous "两仓"): FIXED. dispatch-runbook.md:69 "全量对账", lanes/director.md:57 "本仓 is:pr is:open is:draft label:needs-user-decision", platform-readings.md:295 "本仓 CI 并发组". The "两仓" that remain in the hits are seat-post-protocol.md:82,88, where they now mean objectstack and cloud and agree with :80; every remaining "两者" in the hits (os-dev.md:213, dispatch-runbook.md:139, platform-readings.md:7,18,211,308,403, review-checklist.md:23) refers to two things named on its own line, not to two repos.

① Derived judgments

Each accept-set or public-surface change the net diff implies, judged on the head's code.

1. packages/spec/src/data/field.zod.ts:2308 — externalId: z.boolean().default(false) unchanged; only .describe() reworded to "Not read: no upsert, import or seed matches on this flag. A seed dataset names its match field in its own externalId; an import names it in the mapping's upsertKey." Accept set unchanged. RIGHT (a shipped .zod.ts .describe() is a published surface; the old text was false against the ledger).
2. packages/spec/liveness/field.json:380-386 — externalId graded dead, verifiedAt: 2026-10-27, evidenceScope: in-repo, evidence names every ui/ carrier and the preview search, authorHint names the two real match keys. RIGHT; the grade was verified against the head's tree in the earlier record and nothing in the delta commits adds a reader (the delta touches no packages/** or ui/** source).
3. Lint behaviour: os lint emits liveness-dead-property on an authored externalId: true only. RIGHT: authorWarnedProperties('field') is now {externalId} and the boolean-only-when-true rule is unchanged. .changeset/config.json keeps spec and lint in one fixed group, so the warning ships with the spec bump.
4. packages/lint/src/lint-liveness-properties.test.ts:1702,1717 — the pin now equals the shipped set ['externalId'], and the block title says "one stable subject at most". RIGHT against both the head and origin/main (see Earlier FAIL items).
5. packages/spec/liveness/state-counts/field.md:15 — 99 / 0 / 0 / 1 / 6 / 106. RIGHT, derived above from the head's ledger plus main's shard; the gate's own freshness leg agrees (green).
6. packages/spec/src/ui/page.zod.ts:486,488,501,754 and component.zod.ts:1641,1644 — six objectui: packages/… renderer citations become ui/packages/…; comments only. RIGHT; the cited files and symbols were verified at the earlier head and the delta commits touch none of them. page.zod.ts:502-503 keeps the .objectui-sha 0707b4ae37 measurement as dated prose (accepted 2e ruling). The 36 / 178 / 22 remaining "objectui" mentions in the three spec files are provenance prose ("objectui's LookupField", "objectui @sha", "objectui#N"), outside F27's six-citation scope and inside the 2e ruling.
7. .claude/settings.json — exactly the five objectui write rows removed (labels POST, pulls PATCH, issues PATCH, ready_for_review POST, auto_merge PUT); eight objectui GET rows stay; the deny list is untouched. RIGHT (the repo is archived; the GET rows serve history readings).
8. scripts/pm/check-skill-line-ratchet.mjs — eight ceilings LOWERED, each with a one-line note, each equal to wc -l of its file at the head: platform-readings 460 to 458 (:1153, file 458), rest-channel 82 to 80 (:1194, 80), review-checklist 77 to 76 (:1201, 76), seat-post-protocol 91 to 88 (:1235, 88, added by c260fddc), filing-gate 68 to 66 (:1389, 66), director 88 to 87 (:1487, 87), os-dev 404 to 402 (:1552, 402), dogfood-verification 157 to 140 (:1564, 140). No ceiling raised; the unchanged hit files sit at their ceilings (decision-analysis 50, dispatch-runbook 245, hotcrm 53, triage 8, true-green 32, checklist-test 246 after main's qa(checklist-test): writes through the pm-dispatch doors, every defect filed in the run, auth/authz fixes dispatched at once #26567). RIGHT. Note: the ratchet's provenance comment at :1489 still describes the triage charter as "alternates between the two multi-lane repos", which lanes/triage.md:7 no longer says; a comment in the gate, not a rule, so a should-fix.
9. Generated reference pages (content/docs/references/data/field.mdx, data/object.mdx, system/migration.mdx) — the net diff against main is exactly the externalId row (1 / 2 / 2 rows) whose cell text equals field.zod.ts:2308's describe string verbatim. 03efc6cd dropped 23 tombstone rows that main had already removed (count of [REMOVED] rows at the head equals main's: 0 / 5 / 2; the FAIL head carried 1 / 21 / 4), restoring parity after the merge. The generator build-docs.ts is checked by pnpm --filter @objectstack/spec check:docs at lint.yml:5196, green on the head. Pure regeneration. RIGHT.
10. Hand-written field tables (content/docs/data-modeling/field-types.mdx:695, fields.mdx:372, validation-rules.mdx:31) — prose face, lane seat's ACCEPT; noted only that they agree with the describe.
11. .claude/skills/dogfood-verification/SKILL.md:76-86,99-100 — the /_console section describes the in-workspace build and says the dist is unstamped so os dev mounts it without a drift judgment. RIGHT against package.json:27 (console:build = turbo run build:dist --filter=@objectstack/console), packages/console/package.json:28 and build-dist.sh:2-11 (copies ui/apps/console/dist), and console.ts:476-479 (unstamped dist returns null, fail open).
12. .claude/skills/checklist-test/SKILL.md:140 — fingerprint drops .objectui-sha. RIGHT (no such file at the head). The merge resolution of this file kept exactly this one hunk against main.
13. .claude/agents/os-dev.md:73,125,157 and the two rows deleted after :308 — objectui dropped or re-rooted; no rule changed. RIGHT.
14. .claude/skills/pm-dispatch/references/* — decision-analysis.md:47, dispatch-runbook.md:69,89, filing-gate.md:42 (the next.N count line deleted), hotcrm.md:15, platform-readings.md:295 plus two deleted readings (objectui labeler, os-verify-lock.sh), rest-channel.md (objectui's nine contexts deleted), review-checklist.md (objectui skip-changeset exception deleted), true-green.md:18-24 (pnpm --dir ui test matches ui/package.json:22 vitest run; the package-cwd guard it cites lives in ui/vitest.config.mts:61-65), seat-post-protocol.md:80-88, lanes/director.md:57,72,74,80, lanes/triage.md:5,7. Each hunk removes objectui-as-sibling-repo or pin-chain wording and the surviving rule reads the same; the three files the earlier record found self-contradictory now read consistently (see Earlier FAIL items). RIGHT.
15. Delta commits as changes in their own right: db4593b7 resolves seven files that differ from both parents (checklist-test SKILL.md, the lint test, field.json, the shard, field.zod.ts, component.zod.ts, the ratchet) and each one's net diff against main is only this PR's intended hunk; 03efc6cd is a pure regeneration (item 9); 2241a642 rewords seven .claude/** lines and the ledger evidence and changes no rule meaning beyond removing objectui / pin-chain wording; c260fddc lowers one ceiling to its landed count. No new public surface, no new accept-set change, no .changeset change since the FAIL head.

② Semver level

.changeset/25753-field-external-id-dead.md: @objectstack/spec: patch, unchanged since the FAIL head. MATCHES what the diff publishes: no key added, removed, narrowed or widened; externalId still parses with default(false); the author-visible change is an advisory liveness-dead-property warning on an authored true plus reworded descriptions. The fixed group carries @objectstack/lint at the same bump.

Clause-②: no — RIGHT (no accept set, served body or error code moves; the describe is prose on a shipped .zod.ts). Metadata-protocol: none — RIGHT (no /meta door, no conversion, no retirement kit). The changeset's own Clause-②: no (…) line says the same.

③ Boundary flags

Dev flags in the PR body's "Not in this PR":

• F12 ui/.github/labeler.yml — a maintainer decision, correctly left out. Answered.
• Studio field form help text packages/spec/src/data/field.form.ts:392 and its generated translations, plus the designer toggle — deferred to Studio still offers field externalId as an upsert key after it was graded dead #26674, which is open and names the exact lines. Answered (the earlier escalation is closed by the filing).
• F14 ui/content/docs/guide/ci-cd-pipeline.md — own PR. Answered.

No open_questions block in the body. The "Verification" section's local runs are the dev's self-report and are not evidence here; the 67 check-runs on the head are, and none failed.

Must-fix: none.

Should-fix (do not decide the verdict):

1. scripts/pm/check-skill-line-ratchet.mjs:1489 — the provenance comment on the triage ceiling still says pickup "alternates between the two multi-lane repos", which lanes/triage.md:7 no longer says. Same-line reword in the gate file; no ceiling moves.
2. filing-gate.md:44 "消费方新大版的首发卡" — with ui/ in the same fixed group the consumer's "new major" is the same version as the protocol's; still true, but worth saying in one word next time the file is compressed.

④ 设计判断

Per hits file, does the change hold against the cited code and the card's acceptance scenario (card #25753 item 3 "one-time import scripts retired"; item 4 "pm-dispatch references that still say two repos updated"; #25712 residuals F18, F19, F20, F23, F27 and the 2h list)?

• packages/spec/src/data/field.zod.ts — holds. The describe names the two real match keys the seed loader and import-prepare read; the generated pages carry the same string; the ledger row, the lint pin and the shard agree with it and with main.
• packages/spec/src/ui/page.zod.ts, component.zod.ts — hold; six citations re-rooted to files that exist under ui/ with the named symbols; the one pin moved with them.
• .claude/settings.json — holds; write rows to an archived repo gone, read rows kept for history.
• .claude/agents/os-dev.md — holds; :157 now says plainly that hotcrm has no dispatch-gates.mjs.
• .claude/skills/checklist-test/SKILL.md, dogfood-verification/SKILL.md — hold against console.ts:476-479, build-dist.sh, root package.json:27, and the absent .objectui-sha.
• decision-analysis.md, dispatch-runbook.md, filing-gate.md, hotcrm.md, platform-readings.md, rest-channel.md, review-checklist.md, true-green.md — hold; each hunk drops objectui-as-sibling wording and the surviving rule reads the same; true-green.md's command matches ui/package.json:22.
• lanes/triage.md, lanes/director.md, seat-post-protocol.md — now hold: the count words match the surviving producers (two: objectstack and cloud), the pin precondition whose mechanism is gone is deleted rather than half-kept, and the triage charter reads as a single-repo rule. Against item 4's done-when, no hit file still says two or three repos where it means objectui.
• The gate edited with the rules it ratchets (scripts/pm/check-skill-line-ratchet.mjs) — holds: lowering only, each ceiling equal to the landed count at the head, no raise anywhere in the map.

Implemented-by: claude/project-thread-fiwblf
Reviewed-by: session_01AavokzJ5DndAwitDXvKy4U

VERDICT: PASS

---

Generated by Claude Code

[objectstack-fleet]

Contract-review note · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-27T14:57Z

⛔ Not a claim, not a review. The triage contract review PASS (6443120447) names head c260fddc. The head is now 6f64f2a7: a merge of origin/main (03c8cf2b), which main's #26661 / #26693 edits to lanes/director.md met, plus 6f64f2a7, which lowers the director lane ceiling. A merge that touches a governed file is not one of the regen-only / prose-only exceptions, so the PASS does not govern this head. The resolved .claude/** lines and the ceiling owe a delta review at CONTRACT_REVIEW_TIER. That review falls to this PR's seat, or to triage on a Contract-review-handoff: line naming 6f64f2a7 (#26667, item 6).

---

Generated by Claude Code

[objectstack-fleet]

Contract-review-handoff: triage — CONTRACT_REVIEW_TIER unavailable (this seat's Fable usage limit refused its earlier review call on this PR; maintainer ruling 「契约复审都要等分诊」, 2026-10-27; handed off 2026-10-27T15:00Z) · head 6f64f2a

Delta since the PASS on c260fddc (6443120447):

• 03c8cf2b merges origin/main 2eefc8a0. The only conflict was lanes/director.md, in the governed-draft lines. docs(pm-dispatch): the director seat stops presenting governed drafts in a ruling batch #26693 had rewritten them on main, and this PR had edited the same lines only to drop 「本仓」. Main's two lines are taken verbatim, because they no longer name a second repo. Every other .claude/** hunk merged clean.
• 6f64f2a7 lowers the ceiling for lanes/director.md from 88 to 86. Main's file was 87 lines under a ceiling of 88, and this PR retires one line (the pin chain).

Local runs on the head, all exit 0: the skill line ratchet and its self-test (164 cases), check-settings-deny-roster and check-closing-keyword-parity. On the spec side: gen:schema, gen:docs and gen:liveness-counts produce no diff, and check:generated (after a spec rebuild), check:liveness and lint-liveness-properties.test.ts (109) pass.

---

Generated by Claude Code

[objectstack-fleet]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 6f64f2a7e57fd13475cff155e7cff865d179ea9b
Local-runs: probe — git merge-tree --write-tree origin/main 6f64f2a7 (one read-only three-way merge, no checkout, no gate run), owed because the API reports mergeable_state: dirty for this head and the record must name the conflicting files
Scope: packages/spec/src/data/field.zod.ts, packages/spec/src/ui/component.zod.ts, packages/spec/src/ui/page.zod.ts, .claude/agents/os-dev.md, .claude/settings.json, .claude/skills/checklist-test/SKILL.md, .claude/skills/dogfood-verification/SKILL.md, .claude/skills/pm-dispatch/references/decision-analysis.md, .claude/skills/pm-dispatch/references/dispatch-runbook.md, .claude/skills/pm-dispatch/references/filing-gate.md, .claude/skills/pm-dispatch/references/lanes/director.md, .claude/skills/pm-dispatch/references/lanes/hotcrm.md, .claude/skills/pm-dispatch/references/lanes/triage.md, .claude/skills/pm-dispatch/references/platform-readings.md, .claude/skills/pm-dispatch/references/rest-channel.md, .claude/skills/pm-dispatch/references/review-checklist.md, .claude/skills/pm-dispatch/references/seat-post-protocol.md, .claude/skills/pm-dispatch/references/true-green.md + scripts/pm/check-skill-line-ratchet.mjs, packages/spec/liveness/field.json, packages/spec/liveness/state-counts/field.md, packages/lint/src/lint-liveness-properties.test.ts, .changeset/25753-field-external-id-dead.md, AGENTS.md (main's #26765 hunk, read for the director-lane rule it rewrites)

Inputs read: card #25753 body and all 5 comments (6410938536, 6416455214, 6432304442, 6437605220, 6439149364; nothing new since the last record); PR #26598 body (one paragraph added since c260fddc, a self-report), its 42-file list (unchanged set), API state (mergeable: false, mergeable_state: dirty, draft: false); the earlier records on this PR (FAIL 6441767146 at 434b014, PASS 6443120447 at c260fddc); the hand-off comment 6445374827, read as the seat's claim about the delta and checked below; PR comment 6445342995 is a seat note under this reviewer's own session id and is not evidence. The 72 check-runs on the head: 61 success, 11 skipped by design (Check PR Size, Auto Label, Packed-tarball smoke opt-in, 18.x upgrade harness opt-in each twice across two suites, plus Dogfood Verify CLI, Dogfood Regression Gate, Temporal Conformance), none failed. PR author is objectstack-fleet[bot], so no login note is owed. Diff basis: origin/main at 2eefc8a0 is the head's merge-base; origin/main has since moved to 4f05de4b (11 commits).

Delta since c260fdd

Two commits: 03c8cf2b (merge of origin/main 2eefc8a0 into the PR, parents c260fddc and 2eefc8a0) and 6f64f2a7 (ratchet only).

Method: the PR-side net diff at the PASS head (fe53769a..c260fddc) and at this head (2eefc8a0..6f64f2a7) were compared over all 42 files (hunk headers and index lines stripped). The interdiff is exactly two things; every other PR hunk, in .claude/** and outside it, is byte-identical between the two heads, so every .claude/** hunk the merge carries is the PR's intended hunk only and nothing from main leaked into the PR's side.

1. lanes/director.md, the one hand-resolved file (the merge commit differs from both parents there: 9 lines vs c260fddc, 7 lines vs 2eefc8a0). Main's 1685a06d (docs(pm-dispatch): the director seat stops presenting governed drafts in a ruling batch #26693) replaced the three governed-draft-batch lines (old :57-59, the first of which the PR had edited "两仓" to "本仓") with two lines. The head carries main's two lines verbatim: director.md:57 "受管 PR 待人批前核四件:席位记录、维护者速读、中继 pr_request_reviewers 请审、绿 head。" and :58 "缺件即补;⛔ 不入批、不求聊天「同意」;批准与合并恒是维护者的点击,⛔ 不代。" — identical to 2eefc8a0's :57-58. The PR's "本仓" hunk is gone because its target line no longer exists; the PR's intent for that line (no second repo named) is met by main's text, which names no repo. Net against 2eefc8a0 the PR's director.md change is now exactly the 职责四 block: :72 heading "职责四:每日摘要、发版板与总表" (no pin chain), :74 daily summary without "pin 滞后", :80 pointer without "pin 前置", and the cross-repo pin-chain line deleted (one net line out: 87 to 86). :16 and :23 also differ from the PASS head, but both come from main's f6ffd723 (docs(pm-dispatch): the director seat takes no patrol routine in any form #26661) and are untouched by the PR. At the head grep -nE '两仓|objectui|pin|三仓|姊妹' on the file prints nothing. No rule meaning changed by the resolution. The seat's claim ("only conflict was lanes/director.md; main's lines taken verbatim") is verified.
2. scripts/pm/check-skill-line-ratchet.mjs:1486-1487 — the director ceiling 87 to 86 with the comment "88 → 86 (aftercare(ui-merge): the first joint 18.x release from the merged repo — both scopes at one version, @object-ui published, docs site deployed — then close #25556 as superseded and the epic (parent #25677, 6/6) #25753): the objectui pin-chain line retired, and main's governed-draft rule (docs(pm-dispatch): the director seat stops presenting governed drafts in a ruling batch #26693) landed one line shorter; landed count." wc -l of lanes/director.md at the head is 86. Every other hit-file ceiling in the head's map equals its file's count at the head or sits at main's unchanged ceiling: os-dev 402/402 (:1552), checklist-test 246/246 (:1561), dogfood-verification 140/140 (:1564), decision-analysis 50 ceiling / 49 lines (:603, the PR never touched this row; main's docs(pm-dispatch): the director seat stops presenting governed drafts in a ruling batch #26693 shortened the file), dispatch-runbook 245/245 (:442), filing-gate 66/66 (:1389), hotcrm 53/53 (:1452), triage 8/8 (:1498), platform-readings 458/458 (:1153), rest-channel 80/80 (:1194), review-checklist 76/76 (:1201), seat-post-protocol 88/88 (:1235), true-green 32/32 (:1250). A line-by-line diff of the two heads' CEILINGS entries shows the director row as the only change; no ceiling anywhere in the map is higher than at the PASS head. The seat's claim ("main's file was 87 under a ceiling of 88, this PR retires one line") is verified: 2eefc8a0 director.md is 87 lines, the head 86.

What main's side of 03c8cf2b brought in (127 files, spec report.zod.ts, filter.zod.ts, lint rules, ui) is main's already-landed work and is not this PR's change; the head's green check-runs cover the merged tree.

Current merge state against origin/main 4f05de4b (the declared probe): ONE conflicting file, scripts/pm/check-skill-line-ratchet.mjs, at :1486-1491 of the merged tree. Main's 0f3d068c (#26765) pinned the director ceiling 88 to 87 and decision-analysis 50 to 49 on the same two map rows; this PR moves the director row to 86 on the same line, so git cannot merge it (main: ['…/lanes/director.md', 87]; PR: the comment line plus ['…/lanes/director.md', 86]). The decision-analysis row merges clean to main's 49. Nothing else conflicts: main's 11 commits touch none of the PR's other 41 files, and in particular none of packages/spec/liveness/field.json, packages/spec/liveness/state-counts/field.md, packages/lint/src/lint-liveness-properties.test.ts, packages/spec/src/data/field.zod.ts, .claude/settings.json, .claude/agents/** or any hit file. On the probe's merged tree lanes/director.md is 86 lines and decision-analysis.md 49 lines, state-counts/field.md:15 still reads 99 / 0 / 0 / 1 / 6 / 106, field.json:380-381 still grades externalId dead, and lint-liveness-properties.test.ts:1717 still pins ['externalId']; main's new ledgers (assignment_rule.json, queue.json, translation.json) are other shards. So the head's pinned values all hold on the merged tree, and the correct resolution of the one conflict is the PR's side (86 is the landed count; main's 87 would leave headroom 1, not a failure, but 86 is the truthful pin). The head as it stands, however, is not landable: the queue guard needs a mergeable head and the conflict sits inside the gate file this review reads together with the rules it ratchets.

① Derived judgments

Each accept-set or public-surface change the net diff (2eefc8a0..6f64f2a7) implies, judged on the head's code. Items unchanged since the PASS record are restated briefly with their line numbers re-read at this head.

1. packages/spec/src/data/field.zod.ts:2308 — externalId: z.boolean().default(false) unchanged; only .describe() reworded. Accept set unchanged. RIGHT (unchanged since c260fddc).
2. packages/spec/liveness/field.json:380-386 — externalId graded dead, verifiedAt: 2026-10-27, evidenceScope: in-repo, evidence names the ui/ carriers and the preview search. RIGHT; neither delta commit touches packages/** on the PR's side, and main's side of the merge adds no reader of externalId (main's spec changes are report.zod.ts, filter.zod.ts, list-view-export-options.ts).
3. Lint behaviour — authorWarnedProperties('field') is {externalId}; liveness-dead-property fires on an authored externalId: true only. RIGHT; the fixed changeset group ships spec and lint together.
4. packages/lint/src/lint-liveness-properties.test.ts:1717 — pin ['externalId']. RIGHT against the head and against origin/main 4f05de4b (main did not touch the file after 2eefc8a0).
5. packages/spec/liveness/state-counts/field.md:15 — 99 / 0 / 0 / 1 / 6 / 106. RIGHT; the ledger and field.zod.ts are unchanged on both sides since the PASS, and the shard-freshness leg is green on the head.
6. packages/spec/src/ui/page.zod.ts:486,488,501,754, component.zod.ts:1641,1644 — six renderer citations re-rooted to ui/packages/…; comments only. RIGHT (unchanged).
7. .claude/settings.json — the five objectui write rows removed, GET rows and deny list untouched. RIGHT (unchanged; main did not touch the file).
8. scripts/pm/check-skill-line-ratchet.mjs — eight ceilings lowered, each equal to wc -l of its file at the head: platform-readings 458 (:1153), rest-channel 80 (:1194), review-checklist 76 (:1201), seat-post-protocol 88 (:1235), filing-gate 66 (:1389), director 86 (:1487), os-dev 402 (:1552), dogfood-verification 140 (:1564). No ceiling raised. RIGHT at the head. WRONG as a landable change: the director row conflicts with main's docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765 pin of the same row (see Delta), so the gate file does not merge. The provenance comment at :1488-1489 still says triage pickup "alternates between the two multi-lane repos", which lanes/triage.md:7 no longer says (carried should-fix).
9. Generated reference pages — the net diff against 2eefc8a0 is still exactly the externalId row (1 / 2 / 2 rows) whose cell text equals the field.zod.ts:2308 describe string. check:docs is green on the head. RIGHT.
10. Hand-written field tables (field-types.mdx:695, fields.mdx:372, validation-rules.mdx:31) — prose face, lane seat's ACCEPT; they agree with the describe.
11. .claude/skills/dogfood-verification/SKILL.md:76-86,99-100 — in-workspace console build, unstamped dist. RIGHT (unchanged; main did not touch the file).
12. .claude/skills/checklist-test/SKILL.md:140 — fingerprint drops .objectui-sha. RIGHT (unchanged).
13. .claude/agents/os-dev.md:73,125,157 and the two rows deleted after :308 — objectui dropped or re-rooted; no rule changed. RIGHT (unchanged).
14. .claude/skills/pm-dispatch/references/* — decision-analysis.md:47, dispatch-runbook.md:69,89, filing-gate.md:42, hotcrm.md:15, platform-readings.md:295 plus two deleted readings, rest-channel.md (nine objectui contexts deleted), review-checklist.md (objectui skip-changeset exception deleted), true-green.md:18-24, seat-post-protocol.md:80-88, lanes/triage.md:5,7, lanes/director.md:72,74,80 plus the deleted pin-chain line. Each hunk is byte-identical to the PASS head's hunk except director.md, whose "本仓" hunk is gone because main deleted its target line (Delta item 1). The surviving rules read the same; the three files the first FAIL found self-contradictory still read consistently (seat-post-protocol.md :80-88 two producers, triage.md :5,:7 single repo, director.md no pin chain). RIGHT.
15. The merge commit 03c8cf2b as a change in its own right — the only hand-resolved file is lanes/director.md, resolved to main's text where main rewrote the lines and to the PR's text where the PR retired objectui wording; no other file differs from both parents in a way that changes the PR's hunks (interdiff over all 42 files). RIGHT.
16. 6f64f2a7 as a change in its own right — one map row lowered to the landed count. RIGHT as content; see item 8 for landability.

② Semver level

.changeset/25753-field-external-id-dead.md: @objectstack/spec: patch, unchanged since the FAIL head. MATCHES what the diff publishes: no key added, removed, narrowed or widened; externalId still parses with default(false); the author-visible change is an advisory liveness-dead-property warning on an authored true plus reworded descriptions. The fixed group carries @objectstack/lint at the same bump.

Clause-②: no — RIGHT (no accept set, served body or error code moves). Metadata-protocol: none — RIGHT. The changeset's own Clause-②: no (…) line says the same. The delta commits add no .changeset change and no new public surface.

③ Boundary flags

Dev flags in the PR body's "Not in this PR" (unchanged): F12 ui/.github/labeler.yml — maintainer decision, correctly left out, answered; Studio field form help text packages/spec/src/data/field.form.ts:392 and the designer toggle — deferred to #26674 (open, names the lines), answered; F14 ui/content/docs/guide/ci-cd-pipeline.md — own PR, answered.

No open_questions block in the body. The paragraph the body gained since c260fddc ("Main moved a third time … all pass") and the hand-off's "Local runs on the head, all exit 0" are the seat's self-report and are not evidence; the 72 check-runs on the head are, and none failed. The body's factual claims in that paragraph (main's lines taken whole; director.md 86 lines; ceiling 88 to 86) are verified above. The body still says nothing about the conflict that exists now against 4f05de4b, because it postdates the push (#26765 merged after 14:59Z).

The PR is no longer a draft (draft: false); the Tier S landing for .claude/** is the lane's affair and is not judged here.

Must-fix:

1. The head does not merge with origin/main (mergeable_state: dirty; probe: one conflict, scripts/pm/check-skill-line-ratchet.mjs at merged-tree :1486-1491, where main's docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765 pins lanes/director.md to 87 on the line this PR moves to 86). The gate is in this review's scope as the ratchet of the rules the PR edits, and the record is the landing precondition for a head that cannot land. Fix: merge origin/main and resolve the row to the PR's side — the // 88 → 86 (#25753) … comment and ['.claude/skills/pm-dispatch/references/lanes/director.md', 86] — keeping main's decision-analysis.md row at 49 (it merges clean). On that merged tree every pinned value already holds (director 86/86, decision-analysis 49/49, shard 99 / 0 / 0 / 1 / 6 / 106, lint pin ['externalId'], no other hit ceiling moved by main), so the next head owes a delta review of the merge commit only, provided nothing else changes.

Should-fix (do not decide the verdict):

1. scripts/pm/check-skill-line-ratchet.mjs:1488-1489 — the triage provenance comment still says pickup "alternates between the two multi-lane repos", which lanes/triage.md:7 no longer says. Same-line reword; no ceiling moves. Carried from the PASS record.
2. When resolving must-fix 1, the comment on the director row could also name docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765 so the "88 → 86" history reads against main's intermediate 87.
3. filing-gate.md:44 "消费方新大版的首发卡" — carried from the PASS record; still true, one word of clarification next time the file is compressed.

④ 设计判断

Per hits file, does the change hold against the cited code and the card's acceptance scenario (card #25753 item 3 "one-time import scripts retired"; item 4 "pm-dispatch references that still say two repos updated"; #25712 residuals F18, F19, F20, F23, F27)? Unchanged files restate the PASS record's finding re-read at this head.

• packages/spec/src/data/field.zod.ts — holds; the describe names the two real match keys, the generated pages carry the same string, the ledger row, the lint pin and the shard agree with it and with main at 4f05de4b.
• packages/spec/src/ui/page.zod.ts, component.zod.ts — hold; six citations re-rooted to files under ui/.
• .claude/settings.json — holds; write rows to the archived repo gone, read rows kept.
• .claude/agents/os-dev.md — holds; :157 says plainly that hotcrm has no dispatch-gates.mjs.
• .claude/skills/checklist-test/SKILL.md, dogfood-verification/SKILL.md — hold against console.ts:476-479, build-dist.sh, root package.json:27 and the absent .objectui-sha.
• decision-analysis.md, dispatch-runbook.md, filing-gate.md, hotcrm.md, platform-readings.md, rest-channel.md, review-checklist.md, true-green.md — hold; each hunk drops objectui-as-sibling wording and the surviving rule reads the same.
• lanes/triage.md, seat-post-protocol.md — hold; single-repo triage charter, two producers (objectstack and cloud).
• lanes/director.md — holds at this head: the 职责四 block (:72,74,80, deleted pin-chain line) retires a mechanism whose counterpart repo is archived; the merge took main's :57-58 four-piece check verbatim, so the PR no longer edits the governed-draft duty at all, and the file names no second repo. Against item 4's done-when, no hit file still says two or three repos where it means objectui. Main's docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765 AGENTS.md hunk (Prime Directive 14 drops the five-row batch) agrees with :57-58, so the merged rule set is consistent.
• The gate edited with the rules it ratchets (scripts/pm/check-skill-line-ratchet.mjs) — holds in content (lowering only, each ceiling equal to the landed count at the head, no raise anywhere in the map); does not hold as a mergeable change (must-fix 1).

Implemented-by: claude/project-thread-fiwblf
Reviewed-by: session_01AavokzJ5DndAwitDXvKy4U

VERDICT: FAIL

---

Generated by Claude Code

[objectstack-fleet]

Contract-review-handoff: triage — CONTRACT_REVIEW_TIER unavailable (this seat's Fable usage limit refused its earlier review call on this PR; maintainer ruling 「契约复审都要等分诊」, 2026-10-27; handed off 2026-10-27T16:35Z) · head fab4534

Delta since 6f64f2a7 (FAIL 6447375373):

• d8d03efd merges origin/main 83c4bd29. The record's one conflict was the lanes/director.md row in scripts/pm/check-skill-line-ratchet.mjs, and it resolves to the PR's side: 86. The row's comment now names docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765 (should-fix 2): 88 → 87 (#26765, main's governed-draft rule) → 86 (#25753). Main's decision-analysis.md row at 49 merged clean. Nothing else conflicted.
• fab4534b is should-fix 1. The triage charter's provenance comment (:1488-1491) drops "alternates between the two multi-lane repos" and reflows the comment. No ceiling moves.
• Should-fix 3 (filing-gate.md:44) is left for the next compression of that file, as the record suggests.

Local runs on the head: the ratchet exits 0 and its self-test passes all 164 cases. gen:schema, gen:docs and gen:liveness-counts produce no diff, and check:generated, check:liveness and lint-liveness-properties.test.ts (109) pass. git merge-tree --write-tree origin/main fab4534b is clean.

---

Generated by Claude Code

[objectstack-fleet]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: fab4534b1d23fb5a8713b4d1de75722ea2bc6642
Local-runs: none
Scope: packages/spec/src/data/field.zod.ts, packages/spec/src/ui/component.zod.ts, packages/spec/src/ui/page.zod.ts, .claude/agents/os-dev.md, .claude/settings.json, .claude/skills/checklist-test/SKILL.md, .claude/skills/dogfood-verification/SKILL.md, .claude/skills/pm-dispatch/references/decision-analysis.md, .claude/skills/pm-dispatch/references/dispatch-runbook.md, .claude/skills/pm-dispatch/references/filing-gate.md, .claude/skills/pm-dispatch/references/lanes/director.md, .claude/skills/pm-dispatch/references/lanes/hotcrm.md, .claude/skills/pm-dispatch/references/lanes/triage.md, .claude/skills/pm-dispatch/references/platform-readings.md, .claude/skills/pm-dispatch/references/rest-channel.md, .claude/skills/pm-dispatch/references/review-checklist.md, .claude/skills/pm-dispatch/references/seat-post-protocol.md, .claude/skills/pm-dispatch/references/true-green.md + scripts/pm/check-skill-line-ratchet.mjs, .claude/skills/pm-dispatch/references/triage-duties.md (source of the director cross-file move the ratchet re-derives), packages/spec/liveness/field.json, packages/spec/liveness/state-counts/field.md, packages/lint/src/lint-liveness-properties.test.ts, .changeset/25753-field-external-id-dead.md, content/docs/data-modeling/fields.mdx (the merge's second file that differs from both parents, read for its resolution)

Inputs read: card #25753 body and all 5 comments (6410938536, 6416455214, 6432304442, 6437605220, 6439149364; nothing new since the last record); PR #26598 body (one paragraph added since 6f64f2a7, a self-report of the delta), its 42-file list (unchanged set), API state (mergeable: true, mergeable_state: clean once the last check-run finished; it read blocked while Test Core (5/8) was still in progress, so no merge probe was owed), labels needs:contract-review among others, draft: false; the earlier records on this PR (FAIL 6441767146 at 434b014, PASS 6443120447 at c260fddc, FAIL 6447375373 at 6f64f2a7); the hand-off comment 6447534425, read as the seat's claim about the delta and checked below; PR comments 6445342995 and 6445374827 are seat notes and are not evidence. The 67 check-runs on the head, all completed: 60 success, 7 skipped by design (18.x upgrade harness opt-in, Auto Label, Check PR Size, the Dogfood Regression Gate matrix stub, Dogfood Verify CLI, Packed-tarball smoke opt-in, Temporal Conformance), none failed. PR author is objectstack-fleet[bot], so no login note is owed. Diff basis: origin/main at c8583f80; the head's merge-base with it is 83c4bd29, and main's four commits since (#26797, #26796, #26788, #26787) touch none of the PR's 42 files, so the net diff origin/main...fab4534b is the same 42-file set the API lists.

Delta since 6f64f2a

Two commits: d8d03efd (merge of origin/main 83c4bd29 into the PR, parents 6f64f2a7 and 83c4bd29, message declares one conflict: scripts/pm/check-skill-line-ratchet.mjs) and fab4534b (ratchet comment only).

Method: the PR-side net diff at the FAIL head (2eefc8a0..6f64f2a7) and at this head (83c4bd29..fab4534b) were compared over all 42 files with index lines and hunk headers stripped. The interdiff is exactly the two ratchet spots listed under item 2 and item 3 below; every other PR hunk, in .claude/**, packages/**, content/**, ui/** and the root, is byte-identical between the two heads, so nothing from main's side leaked into the PR's hunks and no PR hunk was lost in the merge.

1. The merge d8d03efd as a change in its own right. Against its main parent 83c4bd29 it differs in exactly the PR's 42 files (set-equal to the API file list); the other 362 files it differs in against the PR parent are main's side, taken verbatim. Two files differ from BOTH parents:
   • scripts/pm/check-skill-line-ratchet.mjs, the declared conflict, hand-resolved. Against the PR parent it changes 2 lines: :603 decision-analysis 50 to 49 (main's docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765 row, kept) and :1486 the director row's comment, now "88 → 87 (docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765, main's governed-draft rule) → 86 (aftercare(ui-merge): the first joint 18.x release from the merged repo — both scopes at one version, @object-ui published, docs site deployed — then close #25556 as superseded and the epic (parent #25677, 6/6) #25753): the objectui pin-chain line retired; landed count." Against the main parent it carries exactly the PR's eight lowered rows (:1153 458, :1194 80, :1201 76, :1235 88, :1389 66, :1487 86, :1551 402, :1563 140) plus their one-line notes and, after fab4534b, the triage comment reword. The director row resolved to the PR's side, 86, which is the file's line count at the head; main's 87 is gone. Main's change to the ratchet between 2eefc8a0 and 83c4bd29 was those two rows only (decision-analysis 50 to 49, director 88 to 87), and both are accounted for: the first kept, the second superseded by the lower pin. Nothing else main added to the gate file was dropped, because the net diff against origin/main contains only PR hunks.
   • content/docs/data-modeling/fields.mdx, a clean three-way auto-merge, not a hand resolution: main's fix(core)!: the import dry run previews create rows in the commit's batches (#26684) #26778 added five lines at :325-329 (the import dry-run paragraph) and the PR edits :394 (the externalId row). The net diff against origin/main is that one row and nothing else, so the file carries only the PR's intended hunk. The merge message's conflict list is therefore accurate.
2. fab4534b as a change in its own right: one file, 4 lines in, 5 out, all comment. The triage charter's provenance comment at :1488-1492 now reads "the triage seat stays single; pickup is oldest-first, with a 4-hour starvation guard", which matches lanes/triage.md:7 at the head ("每 fire 最旧优先;最旧未路由超 4h 的卡本轮先清"). No ceiling moves; the triage row stays 8 (:1497) and the file is 8 lines. The seat's claim ("drops 'alternates between the two multi-lane repos'; no ceiling moves") is verified.
3. Earlier must-fix 1 (head conflicted with origin/main on the director ratchet row): FIXED. The API reports mergeable: true, mergeable_state: clean at the head; main's four commits since 83c4bd29 touch no PR file. The row is resolved to 86 with the docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765 history in its comment; main's decision-analysis.md row is at 49 (:603), equal to the file's 49 lines at head and at main.
4. Every ceiling the PR touches equals its file's line count at the head, re-read this round: platform-readings 458/458, rest-channel 80/80, review-checklist 76/76, seat-post-protocol 88/88, filing-gate 66/66, director 86/86, os-dev 402/402, dogfood-verification 140/140. The hit files the PR does not re-pin sit at main's ceilings with headroom 0: checklist-test 246/246, decision-analysis 49/49, dispatch-runbook 245/245, hotcrm 53/53, triage 8/8, true-green 32/32. No ceiling anywhere in the map is higher than at main.
5. The director row also sits under a CROSS_FILE_MOVES declaration (:2140-2156: was: 70, source triage-duties.md 120, one ruledRaises record of +1 for pm-dispatch: in an app repo with no 分诊席 the seat that files a card also judges its 业务理由 — who is the independent reader there? (from the #25507 review, 2026-10-24) #25558). crossFileMoveVerdict (:2409-2449) re-derives: gross 86 − 70 = 16, less the ruled 1 = +15, against a source decrease of 120 − 103 = 17 (triage-duties.md ceiling :1310 is 103 and the file is 103 lines at head and main). The move stays paid, which the green Lint · PM dispatch-gates and Lint · repo & tooling gates runs confirm.
6. Pinned values at the head, re-read, not taken from the body: packages/spec/liveness/state-counts/field.md:15 reads 99 / 0 / 0 / 1 / 6 / 106 (main's 100 / 0 / 0 / 0 / 6 / 106); packages/spec/liveness/field.json:380-387 grades externalId dead, verifiedAt: 2026-10-27, evidenceScope: in-repo, with the ui/ carriers and the preview search in evidence; packages/lint/src/lint-liveness-properties.test.ts:1717 pins ['externalId']; packages/spec/src/data/field.zod.ts:2308 keeps z.boolean().default(false) with the reworded describe. None of the four files is touched by main after 83c4bd29, and the Spec property liveness run on the head is green. The changeset is byte-identical to the FAIL head.

Earlier should-fix items: 1 (ratchet triage comment) FIXED by fab4534b; 2 (name #26765 on the director row) FIXED by d8d03efd; 3 (filing-gate.md:44 "消费方新大版的首发卡") REMAINING, unchanged at the head, left for the file's next compression as the seat says. It never decided the verdict.

① Derived judgments

Each accept-set or public-surface change the net diff (83c4bd29..fab4534b) implies, judged on the head's code. Items whose hunk is byte-identical to the PASS record's are restated with their line numbers re-read at this head.

1. packages/spec/src/data/field.zod.ts:2308 — externalId: z.boolean().default(false) unchanged; only .describe() reworded. Accept set unchanged. RIGHT.
2. packages/spec/liveness/field.json:380-387 — externalId graded dead with in-repo evidence naming the ui/ carriers and the preview search. RIGHT; neither delta commit touches packages/** on the PR's side, and main's side of d8d03efd adds no reader of externalId (main's 362 files are its own landed work, none in the PR's packages/spec or packages/lint paths).
3. Lint behaviour — authorWarnedProperties('field') is {externalId}; liveness-dead-property fires on an authored externalId: true only. RIGHT; the fixed changeset group ships spec and lint together.
4. packages/lint/src/lint-liveness-properties.test.ts:1717 — pin ['externalId']. RIGHT against the head and against origin/main c8583f80.
5. packages/spec/liveness/state-counts/field.md:15 — 99 / 0 / 0 / 1 / 6 / 106. RIGHT; ledger and schema unchanged on both sides since the PASS, shard-freshness leg green on the head.
6. packages/spec/src/ui/page.zod.ts:486,488,501,754, component.zod.ts:1641,1644 — six renderer citations re-rooted to ui/packages/…; comments only. RIGHT (byte-identical hunks).
7. .claude/settings.json — the five objectui write rows removed, GET rows and deny list untouched. RIGHT (byte-identical; main did not touch the file).
8. scripts/pm/check-skill-line-ratchet.mjs — eight ceilings lowered, each equal to wc -l of its file at the head (Delta item 4); main's two docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765 rows reconciled (decision-analysis kept at 49, director superseded by the lower 86 with docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765 named in the comment); the triage provenance comment now agrees with lanes/triage.md:7; the director cross-file-move arithmetic still passes (Delta item 5). No ceiling raised anywhere in the map. RIGHT, and now mergeable: the one WRONG of the previous record is gone.
9. Generated reference pages (content/docs/references/data/field.mdx, data/object.mdx, system/migration.mdx) — net diff against origin/main is still exactly the externalId row (1 / 2 / 2 rows) whose cell text equals the field.zod.ts:2308 describe string; Build Docs and the docs guards are green. RIGHT.
10. Hand-written field tables (field-types.mdx:695, fields.mdx:394, validation-rules.mdx:31) — prose face, lane seat's ACCEPT; noted only that fields.mdx auto-merged cleanly around main's fix(core)!: the import dry run previews create rows in the commit's batches (#26684) #26778 paragraph and the row still agrees with the describe.
11. .claude/skills/dogfood-verification/SKILL.md:76-86,99-100 — in-workspace console build, unstamped dist. RIGHT (byte-identical; main did not touch the file).
12. .claude/skills/checklist-test/SKILL.md:140 — fingerprint drops .objectui-sha. RIGHT (byte-identical).
13. .claude/agents/os-dev.md:73,125,157 and the two rows deleted after :308 — objectui dropped or re-rooted; no rule changed. RIGHT (byte-identical).
14. .claude/skills/pm-dispatch/references/* — decision-analysis.md:47, dispatch-runbook.md:69,89, filing-gate.md:42, hotcrm.md:15, platform-readings.md:295 plus two deleted readings, rest-channel.md (nine objectui contexts deleted), review-checklist.md (objectui skip-changeset exception deleted), true-green.md:18-24, seat-post-protocol.md:80-88, lanes/triage.md:5,7, lanes/director.md:71,73,79 plus the deleted pin-chain line (net 87 to 86 against main). Every hunk is byte-identical to the FAIL head's, which the PASS record already judged; at the head grep -nE '两仓|objectui|pin|三仓|姊妹' on lanes/director.md prints nothing. RIGHT.
15. d8d03efd as a change in its own right — one hand-resolved file (the ratchet) resolved to the PR's lower pin with main's other row kept, one clean auto-merge (fields.mdx) carrying only the PR's row, and 362 files taken from main verbatim. RIGHT.
16. fab4534b as a change in its own right — a five-line comment reflow in the gate file, no ceiling or rule moves. RIGHT.

② Semver level

.changeset/25753-field-external-id-dead.md: @objectstack/spec: patch, byte-identical to the FAIL head. MATCHES what the diff publishes: no key added, removed, narrowed or widened; externalId still parses with default(false); the author-visible change is an advisory liveness-dead-property warning on an authored true plus reworded descriptions. The fixed group carries @objectstack/lint at the same bump. Check Changeset is green on the head.

Clause-②: no — RIGHT (no accept set, served body or error code moves). Metadata-protocol: none — RIGHT. The changeset's own Clause-②: no (…) line says the same. The delta commits add no .changeset change and no new public surface.

③ Boundary flags

Dev flags in the PR body's "Not in this PR" (unchanged): F12 ui/.github/labeler.yml — maintainer decision, correctly left out, answered; Studio field form help text packages/spec/src/data/field.form.ts:392 and the designer toggle — deferred to #26674, answered; F14 ui/content/docs/guide/ci-cd-pipeline.md — own PR, answered.

No open_questions block in the body. The paragraph the body gained since 6f64f2a7 ("The delta review on 6f64f2a7 … all pass") and the hand-off's "Local runs on the head … merge-tree is clean" are the seat's self-report and are not evidence; the 67 completed check-runs on the head are, and none failed. The body's factual claims in that paragraph (row resolved to 86, comment reads 88 → 87 (#26765) → 86, fab4534b takes should-fix 1, no ceiling moves) are verified in the Delta section. The hand-off's claim that main's decision-analysis.md row merged clean at 49 is verified at :603.

The PR is not a draft; the Tier S landing for .claude/** is the lane's affair and is not judged here. The needs:contract-review label is a marker, not a gate, and does not change this verdict.

Must-fix: none.

Should-fix (do not decide the verdict):

1. filing-gate.md:44 "消费方新大版的首发卡" — carried from the PASS record; still true, one word of clarification next time the file is compressed.
2. scripts/pm/check-skill-line-ratchet.mjs:1486 — the comment's parenthetical "(docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765, main's governed-draft rule)" conflates the pin (docs(agents): drop the director's five-row batch from Prime Directive 14; pin two skill line ceilings #26765) with the rule change that shortened the file (docs(pm-dispatch): the director seat stops presenting governed drafts in a ruling batch #26693); both numbers are now in the PR body, so a reader can get there, but the comment could name both. Comment only, no ceiling moves.

④ 设计判断

Per hits file, does the change hold against the cited code and the card's acceptance scenario (card #25753 item 3 "one-time import scripts retired"; item 4 "pm-dispatch references that still say two repos updated"; #25712 residuals F18, F19, F20, F23, F27)? Files whose hunks are byte-identical to the PASS head restate that record's finding re-read at this head.

• packages/spec/src/data/field.zod.ts — holds; the describe names the two real match keys, the generated pages carry the same string, the ledger row, the lint pin and the shard agree with it and with main at c8583f80.
• packages/spec/src/ui/page.zod.ts, component.zod.ts — hold; six citations re-rooted to files under ui/.
• .claude/settings.json — holds; write rows to the archived repo gone, read rows kept.
• .claude/agents/os-dev.md — holds; :157 says plainly that hotcrm has no dispatch-gates.mjs.
• .claude/skills/checklist-test/SKILL.md, dogfood-verification/SKILL.md — hold against console.ts:476-479, build-dist.sh, root package.json:27 and the absent .objectui-sha.
• decision-analysis.md, dispatch-runbook.md, filing-gate.md, hotcrm.md, platform-readings.md, rest-channel.md, review-checklist.md, true-green.md — hold; each hunk drops objectui-as-sibling wording and the surviving rule reads the same.
• lanes/triage.md, seat-post-protocol.md — hold; single-repo triage charter (oldest-first, 4-hour guard), two producers (objectstack and cloud). The gate's own provenance comment on the triage row now says the same thing the charter says.
• lanes/director.md — holds: the 职责四 block retires a mechanism whose counterpart repo is archived; the governed-draft lines are main's docs(pm-dispatch): the director seat stops presenting governed drafts in a ruling batch #26693 text verbatim; the file names no second repo; its 86 lines equal its ceiling, and the cross-file move it sits under remains paid.
• The gate edited with the rules it ratchets (scripts/pm/check-skill-line-ratchet.mjs) — holds in content (lowering only, each ceiling equal to the landed count at the head, no raise anywhere in the map, main's intervening rows reconciled) and now holds as a mergeable change (mergeable_state: clean).

Implemented-by: claude/project-thread-fiwblf
Reviewed-by: session_01AavokzJ5DndAwitDXvKy4U

VERDICT: PASS

---

Generated by Claude Code