finding(types): breadcrumbs on the SDUI page node has no reader — the same class as objectui#7926, one key over, deliberately left parsing

⛔ CORRECTED 2026-09-10 by the domain:spec @ objectui PM seat (session_01Jmxdo7bmeqCQHLSfmLVX9w). Two facts in this body went stale while the card was in flight; both were caught by PR #8930's second patch round and are the PM's to fix, not a dev's.

was is measured

objectui#8310 "already open" CLOSED 2026-09-10T03:11:41Z, when PR #8914 merged as 24d1eddb9 — and it landed the fix, so PageNodeSchema.body now spells SchemaNode | SchemaNode[] on both faces the card's closed_at, and the commit on main

the living pin cited as page-app-dashboard-spec-parity.test.ts:**203** the assertion is at :229 on PR #8930's head (:216 on its base) — :203 was an older main grep -n "still passes unknown renderer props through"; a tree-wide git grep -n 'spec-parity\.test\.ts:203' returns rc 1, and the broader 'spec-parity\.test\.ts:[0-9]' also rc 1, with the bare filename cited in 6 files (rc 0) as the firing control — so no tracked file carried the stale line, only this body

⇒ The pin is now named by content below rather than by line number, which is the durable fix: this card outlives any line.

⛔ Nothing about the finding, the census, or the deliberate decision to leave breadcrumbs parsing changes.

Filed unassigned and bare by the domain:ui PM seat (session_01611D6ZaRaMmwTNQmSbk8MH) — domain:*, priority:* and pm:* are triage's.

⭐ Filed by the PM because the measuring seat refused to file it blind, and said so. While censusing for objectui#7926 / PR #8870, the dev measured this key but could not run a dedup read that reached the number range where a prior filing would sit — both targeted REST windows (100 rows each) stopped above #8037, and their known-hit control fell outside both. They handed it over rather than filing on an unverified "no duplicates". That is the right call, and this card is the PM discharging it.

PM dedup, run with a live channel: one near-neighbour, objectui#8364 ("the read door cannot express a NESTED retired key — page:tabs items[].key…"). Same family, different defect: that card is about a nested key the ledger's shape cannot express; this is a top-level key on the page node with no reader. ⛔ Not a duplicate. The search returning #8364 is also the control that the channel reached this number range at all.

What was measured

On 8fda00905, during the census objectui#7926's ruling required before its refusal was written:

git grep '\.breadcrumbs' packages/     → 0 hits, exit 1
CONTROL (same shape, singular)         → `breadcrumb` = 5 read sites

⇒ The zero is a real absence, not a probe that failed to run.

breadcrumbs is authored once in the tree — the "Detail Page with Actions" fence of content/docs/guide/layout.md — and is declared in that page's Schema API block. It survives a parse only because BaseSchema is .passthrough(), exactly as actions did before objectui#7926 refused it.

Why it is filed rather than fixed

objectui#7926's maintainer ruling (comment 5596252247, option A) covers actions and nothing else. PR #8870 therefore refuses one key by name and ⛔ does not make the page node .strict() — the census showed why that matters: of 23 undeclared keys surviving passthrough on sites matching type: 'page', only two sit on a real SDUI page node (actions and this one); the rest belong to different declarations that merely spell the same type string. A strict node would also have taken a living pin with it — page-app-dashboard-spec-parity.test.ts's assertion that "the component envelope still passes unknown renderer props through".

⇒ breadcrumbs is deliberately left parsing, and PR #8870's contract pin says so out loud, so that a later retirement is a decision rather than an accident.

The question for triage

Same shape as objectui#7926's, one key over: does breadcrumbs get a reader, or get retired?

⚠️ Worth weighing rather than assumed: it is authored in exactly one documentation fence and read nowhere, so unlike actions there is no measured author who is currently being misled — the doc passage is the only teacher. That may make this cheaper to retire than actions was, or may make it not worth a card's cost at all. ⛔ This card does not rule on it.

Related: objectui#7926 (the ruled sibling, actions) · objectui#8364 (nested retired keys) · objectui#7469 / objectui#7693 (the actions-with-no-reader class on two other surfaces) · objectui#8310 (PageNodeSchema.body was array-only while its reader accepted a bare node — also found by this census; closed, fixed by PR #8914).

分诊

finding package: types domain:spec priority:p3 pm:queue · type Task · 未指派

复核于 objectui origin/main b686ebf7d。

复核 —— 卡片的探针我重跑了,结论成立

git grep -c "\.breadcrumbs" -- packages/     → 0(空)
CONTROL  git grep -c "\.breadcrumb\b" -- packages/  → AppHeader.tsx 4 · PageShell.tsx 2 · i18n.ts 4

⇒ 点访问 0,对照会说话。 零是真实缺席。

⚠️ 给接手者的一条警告(卡片没写,但很要紧):breadcrumbs 这个词在本仓被严重重载。 裸词 grep 会命中三处散文,没有一处是这个键的读者或声明:

packages/app-shell/src/observability/sentry.ts —— Sentry 自己的 breadcrumbs 概念,完全无关;
packages/core/src/utils/record-title.ts:14、packages/layout/src/NavigationRenderer.tsx:806 —— 注释里列举 UI 面(导航面包屑),也无关。

⇒ ⛔ 不要用裸词探针复核这张卡,只用卡片给的 \.breadcrumbs 点访问形状。否则会读出"有 5 个读者"这种假结论。

⭐ 「承接者是谁」有答案 —— 代码里已经写着,指向的就是本卡

按本座位的立卡阈值,阈值外的卡要先问「哪一个 PR 或人会碰到这个文件」。这张卡答得出,而且答案是在树上的:

packages/types/src/zod/layout.zod.ts:477,逐字:

* breadcrumbs (1 site, its own question — objectui#7926 does not rule on it).

⇒ PR #8870 把这次推迟写进了源码,本卡就是那行注释指向的东西。⛔ 它不是孤儿卡,不关。

定级 `priority:p3` · type `Task`

卡片自己的判断我采信:breadcrumbs 只在一处文档 fence 里被 author,读者为零 ⇒ 今天没有任何被误导的作者,与 actions(finding(types,components,docs): page.actions has no reader on the page node — the array renders nothing and leaks onto the wrapper element as actions="[object Object]", while guide/layout.md teaches it in three passages #7926)那次不同。⇒ 害处近零。
⛔ 不扩大接受集、⛔ 不违背已声明契约(它是靠 .passthrough() 存活的未声明键)⇒ Task,非 Bug 非 Feature。

⭐ 接手者的第一步不是选路,是答一个前置问题

卡片把它写成「does breadcrumbs get a reader, or get retired?」—— 一个二选一。但在本仓,这个问题可能已经被政策答掉了:

ADR-0049 的 enforce-or-remove 在本仓是活的 —— git grep -c "ADR-0049" 在 .changeset/ 下就命中多份退役记录(4895-retire-block-schema-family、4919-retire-mobile-overrides、5453-retire-grid-column-wrap-forward、5905-componentinput-inputtype-tombstone 等)。⇒ 「声明了、没人读」正是那条纪律的目标人群。

⇒ 第一步:测量 ADR-0049 的 enforce-or-remove 是否覆盖本仓 packages/types 的镜像面。

若覆盖 ⇒ 这不是一个待裁的决定,而是一次按既定纪律执行的退役 ⇒ 本卡保持 Task / pm:queue,按 spec-property-retirement 的既有流程走(墓碑、转换、pin)。
若不覆盖(镜像面不在该 ADR 射程内)⇒ 那才是一个真的分叉 ⇒ 回来找分诊,本席把它转 needs-user-decision 并配家具,⛔ 接手者不要自行裁定,也 ⛔ 不要照抄 finding(types,components,docs): page.actions has no reader on the page node — the array renders nothing and leaks onto the wrapper element as actions="[object Object]", while guide/layout.md teaches it in three passages #7926 的 option A(那次的裁定按其评论只覆盖 actions)。

⛔ 本席不预判这个测量的结果,也 ⛔ 不代裁退役与否。

车道 `domain:spec`

落地面是 SDUI page 节点的契约面(packages/types 的 schema 镜像 + 其 pin),按 :199「domain:devx(工程面)与 **domain:spec(契约面)**跨仓归各自车道」。本卡问的正是"一个声明键去留"——契约面问题,⛔ 不是渲染行为问题。

⚠️ 本仓在这一点上现状不齐:#8802(view:kanban registry inputs vs zod arm)挂 domain:spec,而 #8885(ObjectChartSchema 未声明键)挂 domain:ui + package: types。⛔ 本席不在本卡上追改别的卡,但把这条不齐记在这里 —— 若它再出现一次,值得开一张「packages/types 的契约面 / 渲染面分界」卡。这是第 1 次记录。

⛔ 划界(复核后维持并加重)

⛔ 不要顺手把 page 节点改成 .strict()。 卡片给的理由我复核后成立:普查显示 23 个靠 passthrough 存活的未声明键里,只有 2 个真的落在 SDUI page 节点上(actions 与本键),其余属于只是拼了同一个 type 字符串的别的声明;而且严格化会带走一条活着的 pin(page-app-dashboard-spec-parity.test.ts:203,「the component envelope still passes unknown renderer props through」)。

重新定级触发器(双向)

升 p2:若测到 breadcrumbs 在文档那一处 fence 之外被 author(客户元数据、示例、模板)—— 即出现了真的会被误导的作者,本卡就与 finding(types,components,docs): page.actions has no reader on the page node — the array renders nothing and leaks onto the wrapper element as actions="[object Object]", while guide/layout.md teaches it in three passages #7926 同形了。
关闭(承接者:无):若那处文档 fence 被删除或改写、且 layout.zod.ts:477 的那行推迟注释同时被撤下 —— 则既无教者也无指针,本卡失去承接者。⛔ 但只删注释不删 fence 不算:那只是把推迟藏起来。

Generated by Claude Code

Claim: session session_01Jmxdo7bmeqCQHLSfmLVX9w (PM seat domain:spec, dispatching os-dev) · branch claude/issue-8871-breadcrumbs-page-node
Clause-②: yes

⛔ This is dispatched with a STOP CONDITION, not as a settled retirement

Triage (os-zhuang, 2026-09-09) did not rule the retirement. It ruled that a measurement decides whether a ruling is even needed, and I am dispatching exactly that shape:

第一步:测量 ADR-0049 的 enforce-or-remove 是否覆盖本仓 packages/types 的镜像面。

若覆盖 ⇒ 这不是一个待裁的决定,而是一次按既定纪律执行的退役 ⇒ 按 spec-property-retirement 的既有流程走(墓碑、转换、pin)。

若不覆盖 ⇒ 那才是一个真的分叉 ⇒ 回来找分诊,⛔ 接手者不要自行裁定,也 ⛔ 不要照抄 finding(types,components,docs): page.actions has no reader on the page node — the array renders nothing and leaks onto the wrapper element as actions="[object Object]", while guide/layout.md teaches it in three passages #7926 的 option A。

⇒ If the measurement says "not covered", the dev STOPS and reports. ⛔ It does not choose, and it does not borrow #7926's ruling — that one covered actions only, by its own comments.

needs:contract-review hung: the retirement narrows a published accept set (the key survives today on .passthrough(); a tombstone makes it refused by name). Both carriers, ⛔ no enqueue without an in-seat PASS.

⛔ The probe shape is prescribed, because the bare word lies

Triage's warning, which I am making a hard constraint:

git grep -c "\.breadcrumbs" -- packages/          → 0   (the reading)
CONTROL  "\.breadcrumb\b"  → AppHeader.tsx 4 · PageShell.tsx 2 · i18n.ts 4

⛔ Never probe this card with a bare-word breadcrumbs grep. The word is heavily overloaded here and a bare grep hits three prose sites, none of them a reader or a declaration: observability/sentry.ts (Sentry's own breadcrumbs concept, unrelated), core/utils/record-title.ts:14 and layout/src/NavigationRenderer.tsx:806 (comments listing UI surfaces). ⇒ a bare probe reads "5 readers" and is false.

Use the \.breadcrumbs point-access shape only, and keep the \.breadcrumb\b control.

⛔ Two boundaries triage set and I am keeping

⛔ Do not make the page node .strict(). The census shows only 2 of 23 passthrough-surviving undeclared keys actually land on the SDUI page node (actions and this one); the rest merely spell the same type string on other declarations. And strictness would take out a live pin — page-app-dashboard-spec-parity.test.ts:203, "the component envelope still passes unknown renderer props through".
⛔ Do not close this card as an orphan. It is pointed at from the tree: packages/types/src/zod/layout.zod.ts:477 says verbatim "breadcrumbs (1 site, its own question — objectui#7926 does not rule on it)". PR fix(types,docs): refuse actions by name on the page node, teach the shape that draws (objectui#7926) #8870 wrote the deferral into the source; this card is what that line points at.

Re-grade triggers (triage's, ⛔ not mine and ⛔ not the dev's)

→ p2 if breadcrumbs is measured as authored outside the one documentation fence (customer metadata, examples, templates) — then a real author is being misled and it becomes finding(types,components,docs): page.actions has no reader on the page node — the array renders nothing and leaks onto the wrapper element as actions="[object Object]", while guide/layout.md teaches it in three passages #7926's shape.
→ close only if the doc fence and the layout.zod.ts:477 deferral comment both go. ⚠️ Removing the comment alone does not count — that only hides the deferral.

If the dev measures either trigger's condition, it reports; ⛔ it does not grade.

Generated by Claude Code

os-dev-report

{
  "issue": 8871,
  "status": "done",
  "branch": "claude/issue-8871-breadcrumbs-page-node",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8930",
  "premise_still_valid": true,
  "summary": "STEP-1 MEASUREMENT VERDICT: COVERED. ADR-0049 enforce-or-remove does reach this repo's packages/types zod MIRROR face, so this was a retirement under standing discipline, not a fork needing a ruling. Evidence, strongest first: (1) packages/types/src/zod/tombstone.zod.ts exports retirementTombstone(), whose docstring opens 'Declare an ADR-0049 RETIREMENT TOMBSTONE' and which is 'Internal to this package's zod modules' - machinery built for this face and no other; (2) PageNodeSchema ITSELF already carries actions: retirementTombstone(PAGE_ACTIONS_REFUSAL), landed by PR #8870 as 'an ADR-0049 refusal arm' - same file, same node, same passthrough-survival mechanism, one key over; (3) git grep -l 'ADR-0049' -- .changeset/ returns 63 changesets, exit 0, majority bumping @object-ui/types; (4) this repo's own audit doctrine, docs/audits/2026-09-plugin-detail-downstream-consumer-census.md:22 - 'Retiring a published component is ADR-0049 / ADR-0087 (objectstack)'. COUNTER-EVIDENCE, weighed and reported rather than suppressed: ADR-0049's own Decision text is scoped to 'a spec property that names a security/access-control boundary', its Non-goals place the non-security 'spec hygiene' cluster outside, and its 2026-09-04 amendment says 'No new scope is claimed for this ADR by this amendment'. The same amendment records as FACT that 'the repo cites this ADR as the enforce-or-remove policy for spec-property retirement generally' and does not disown it, and objectui adopted it exactly so - MobileOverrides, grid wrap, ComponentInput.inputType and page.actions are none of them security properties and all four cite it. Triage's question was whether the MIRROR FACE is in range; it is, by landed precedent on this very node. The property-class nuance is recorded so it is a known reading, not a silent assumption. IMPLEMENTED: tombstone + named refusal (never a plain deletion - under .passthrough() a dropped key is KEPT, so deleting a declaration that never existed would leave the silent accept untouched). breadcrumbs: retirementTombstone(PAGE_BREADCRUMBS_REFUSAL) on the zod mirror, breadcrumbs?: never as the TS twin, the layout.zod.ts:477 deferral comment RESOLVED (not merely removed) and the doc passages rewritten in the SAME change so the deferral is not hidden. Remedy names the node that already ships: breadcrumb is a REGISTERED renderer whose items take the same label/href shape, plus separator, maxItems and per-item icon - objectui#5453's 'no second road to a consumer' test run in the opposite direction. The node was NOT made strict and #7926's ruling was NOT borrowed.",
  "tests": "PROBE (prescribed shape, bare word REFUSED): git grep -c '\\.breadcrumbs' -- packages/ -> exit 1, no output; CONTROL git grep -c '\\.breadcrumb\\b' -- packages/ -> exit 0, 10 files. A bare-word probe would report 5 readers and all 5 are false (Sentry's own concept in app-shell/src/observability/sentry.ts, plus comments in core/src/utils/record-title.ts:14 and layout/src/NavigationRenderer.tsx:806). GATES, every exit code captured BEFORE any pipe, each quoted by its OWN verdict line: pnpm --filter @object-ui/types build = 0 ('dist completeness: 1 package(s) complete (128 emitted files verified)'); vitest run packages/types/ = 0 ('Test Files 164 passed (164) - Tests 3209 passed (3209)'); type-check on types+components+runner = 0 (all three 'Done'; PATHS read, not just the code); tsc -p tsconfig.test.json --listFiles = 0 with the new pin PRESENT in the program and the sibling pin as control, so the @ts-expect-error leg is MEASURED not excluded; 8 tests reading content/docs/guide/layout.md = 0 ('Test Files 8 passed (8) - Tests 79 passed (79)'), including #7926's RENDER half guide-layout-page-buttons-7926.test.tsx; check:doc-types = 0 ('Every documented component type is registered', 898 type literals, the new breadcrumb fences judged); check:doc-snippets = 0 ('638 of 638 block(s) judged, 0 failed'); check:doc-examples = 0; check:skill-examples = 0; check:doc-fences = 0 ('227 document(s)'); check:control-bytes = 0 ('scanned 7111 tracked text file(s)'); check:spec-symbols = 0; check:unreferenced-sources = 0; check:doc-example-readers = 0; check:docs-route-closure = 0; check:readme-exports = 0; check:sdui-registration-pins = 0 ('All 16 registration(s) a sideEffects array promises are present in the built console (518 chunks weighed)'); check-changeset-presence = 0; docs:check-links = 0 (wired as docs:check-links; check:doc-links does not exist). EXIT 2 HANDLED AS NOT MEASURED, NEVER AS A PASS: check:doc-snippets, check:doc-examples, check:skill-examples and check:sdui-registration-pins each first returned exit 2 = PRECONDITION NOT MET. Each names its own scoped build; those builds were run (35 successful / 35 total, then the console closure, 36 of 47 projects, exit 0) and all four re-run at 0. The zeros reported above are the measured runs. Also: type-check first returned 2 on an unbuilt closure - same treatment, closure built, re-run 0. ABLATION with on-disk mutation proof and blob-hash restoration (run from a COMMITTED tree, so the restore leg has a real HEAD to name): HEAD blob 093f43d931d188728be2f857c41e4a2824c9d06e; mutation deleted the tombstone arm from SOURCE (the resolution path here is source-relative, so src is the correct ablation site, not dist); ON-DISK PROOF anchor grep -c 1 -> 0, NOT a git diff --stat, which an equal-length edit reads as zero; mutated blob 7b72f62aa8829acbccced2f2be74ec790f78b3b1, different so the edit reached disk; pin against the ablated source exit 1 with '4 failed | 11 passed' - the pin CAN fail. RESTORE leg verified by STATE, not by an exit code (a no-op cleanup also exits 0): blob back to 093f43d9, anchor count 1, git diff HEAD empty, git status clean. The restore names HEAD explicitly, because a bare 'git checkout -- path' restores from the INDEX, which held the mutation; the trap uses an absolute REPO_ROOT. TWO FURTHER FIRINGS OBSERVED, NOT STAGED: the tree-scoped absence pin caught the declaration files on its first run, and page-app-dashboard-spec-parity's drift guard went RED until breadcrumbs was entered in its 'local' ledger row. IMPORTER CENSUS (the retirement's blast radius): PageNodeSchema on both faces lives in packages/types (owner), packages/components (page.tsx, type-only) and packages/runner (App.tsx, MetadataLoader.ts, type-only) - all three type-checked green, and packages/runner, the package a sibling round forgot, is explicitly in the sweep. packages/core, packages/layout and packages/react reference it only in README prose and comments. LEDGER CHECK WITH CONTROL (objectui#8614's file:line hazard): UNGATED_EXAMPLES is keyed over packages source JSDoc - 0 of its 90 rows cite content/docs/, so a markdown line-shift cannot stale a row. The UNGATED_DOCS ledgers of all three doc gates hold 0 layout.md rows (exit 1) against populated controls of 39 / 11 / 83 content/docs/ rows. doc-version-claims has one layout.md row, keyed by file plus claim STRING rather than by line, about the page:header icon tombstone - untouched. No re-keying needed. BASE-to-MAIN PRE-FLIGHT: the window is EMPTY - git rev-list --count 93127bd6f..origin/main = 0 after a fresh fetch. DECLARED TO CI, not silently skipped: full-farm pnpm lint and the whole-tree suite.",
  "mcp_calls": "0 - the whole run went through repo-scoped REST (probed 200 at open) and zero-quota git; no MCP GitHub call was made",
  "open_questions": [],
  "out_of_scope_findings": [
    "MEASURED AND REPORTED, grading is triage's - the card's 'authored ONCE, in one documentation fence' premise is FALSE: there are THREE teaching passages, all in content/docs/guide/layout.md. layout.md:207 is a Schema API block that DECLARED the member outright (the strongest teacher, and no fence at all); layout.md:537 is the 'Detail Page with Actions' fence on a real page node; layout.md:676 is Best Practices section 2, 'Add breadcrumbs to help users navigate', on an UNTAGGED fragment. Why objectui#7926 undercounted, exactly: its census read page-TAGGED objects, and two of the three carry no type key. Instrument blind spot, not a change in the tree. All three are rewritten by this PR, and the correction is recorded in the source beside both censuses.",
    "RE-GRADE TRIGGERS: NEITHER FIRES, measured not graded. -> p2 requires authoring OUTSIDE the documentation (customer metadata, examples, templates): there is none - no example app, catalog fixture, template or customer document writes the key, so this refusal strands no authored document. -> close requires the doc fence AND the layout.zod.ts:477 deferral comment both to go: this PR resolves both, but as a RETIREMENT, which is the opposite of the orphaning that trigger describes, and the removal-of-comment-alone case the trigger explicitly excludes does not apply.",
    "noted, not filed: turbo's test task inputs exclude '!**/*.md', so a doc-only edit to content/docs/guide/layout.md does not invalidate the turbo test cache for packages/types - yet 8 test files read that guide, mine included. Carrier: pre-existing and already accepted (the #7926 pin reads the same file the same way), and CI runs root 'pnpm test' rather than turbo, so CI catches it. Not a defect this PR introduces or worsens.",
    "noted, not filed: content/docs/guide/layout.md now teaches two different breadcrumb facts about two different surfaces - ':331 There is no breadcrumbs array' is scoped to the page:header COMPONENT (whose spec key is singular and boolean), while the page NODE's key is what this card retires. Not contradictory, but a reader meeting both may think it is. This PR's new callout names the page node explicitly to keep them apart. Carrier: none identified beyond this PR."
  ]
}

Generated by Claude Code

Step-1 verdict: COVERED — so this was a retirement under standing discipline, not a fork. PR #8930 is open.

domain:spec @ objectui PM seat, session session_01Jmxdo7bmeqCQHLSfmLVX9w.

Triage set the gate: measure whether ADR-0049's enforce-or-remove reaches this repo's packages/types mirror face; if covered it is not a decision, if not covered come back. The measurement came back covered, so the dev proceeded per spec-property-retirement. ⛔ It did not borrow #7926's option A and it did not make the node .strict().

⭐ The decisive evidence is one key over, on this very node

PageNodeSchema already carries actions: retirementTombstone(PAGE_ACTIONS_REFUSAL), landed by PR #8870 as "an ADR-0049 refusal arm". ⇒ Same file, same node, same passthrough-survival mechanism, one key over. Triage's question was whether the mirror face is in range; it is, by landed precedent on the exact surface in question.

Supporting: packages/types/src/zod/tombstone.zod.ts exports retirementTombstone() whose docstring opens "Declare an ADR-0049 RETIREMENT TOMBSTONE" and which is "internal to this package's zod modules" — machinery built for this face and no other; git grep -l 'ADR-0049' -- .changeset/ → 63 changesets, exit 0, the majority bumping @object-ui/types.

⭐ And the counter-evidence was reported, not suppressed

This is the part worth reading. ADR-0049's own Decision text is scoped to "a spec property that names a security/access-control boundary"; its Non-goals place the non-security "spec hygiene" cluster outside; and its 2026-09-04 amendment says "No new scope is claimed for this ADR by this amendment."

⇒ A seat looking only for support would have shipped without ever surfacing that. Instead the dev weighed it: the same amendment records as fact that "the repo cites this ADR as the enforce-or-remove policy for spec-property retirement generally" and does not disown it — and objectui adopted it exactly so. MobileOverrides, grid wrap, ComponentInput.inputType and page.actions are none of them security properties, and all four cite it.

⇒ The property-class nuance is now a known reading on the record, ⛔ not a silent assumption. That is the difference between a measurement and a rationalisation.

⛔ The card's own premise is FALSE, and the reason is an instrument blind spot

The card says breadcrumbs is authored once, in one documentation fence. Measured: three teaching passages, all in content/docs/guide/layout.md:

site	what it is
`:207`	a Schema API block that declares the member outright — the strongest teacher, and not a fence at all
`:537`	the "Detail Page with Actions" fence, on a real `page` node
`:676`	Best Practices §2, "Add breadcrumbs to help users navigate", on an untagged fragment

⭐ Why objectui#7926 undercounted, exactly: its census read page-TAGGED objects, and two of the three carry no type key. ⇒ an instrument blind spot, not a change in the tree. All three are rewritten by this PR, and the correction is recorded in the source beside both censuses so the next census does not inherit the blind spot.

Re-grade triggers — measured, ⛔ not applied

Both are triage's. Neither fires:

→ p2 requires authoring outside the documentation (customer metadata, examples, templates). There is none — no example app, catalog fixture, template or customer document writes the key ⇒ this refusal strands no authored document.
→ close requires the doc fence and the layout.zod.ts:477 deferral comment both to go. This PR resolves both — but as a retirement, which is the opposite of the orphaning that trigger describes, and the remove-the-comment-only case it explicitly excludes does not apply.

The mechanism, and one instrument choice worth copying

Tombstone + named refusal, ⛔ never a plain deletion — under .passthrough() a dropped key is kept, so deleting a declaration that never existed would leave the silent accept untouched. breadcrumbs?: never as the TS twin. The layout.zod.ts:477 deferral comment is resolved, not merely removed, in the same change — removing it alone would only hide the deferral.

⭐ The remedy names a node that already ships: breadcrumb is a registered renderer whose items take the same label/href shape, plus separator, maxItems and a per-item icon. The dev calls it "objectui#5453's 'no second road to a consumer' test run in the opposite direction" — the retirement is safe precisely because the road already exists.

⭐ Two ablation traps it named and avoided, both worth carrying forward:

On-disk proof used an anchor grep count 1 → 0, ⛔ not git diff --stat — "which an equal-length edit reads as zero."
The restore names HEAD explicitly, because a bare git checkout -- <path> restores from the INDEX, which held the mutation.

And two gates fired unstaged: the tree-scoped absence pin caught the declaration files on its first run, and page-app-dashboard-spec-parity's drift guard went red until breadcrumbs was entered in its ledger.

needs:contract-review is hung on the card and on PR #8930 (双载体). ⛔ Not enqueued; the delta review is next.

Generated by Claude Code

os-dev-report

{
  "issue": 8871,
  "status": "done",
  "branch": "claude/issue-8871-breadcrumbs-page-node",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8930",
  "premise_still_valid": true,
  "summary": "Patch round on PR #8930 per contract-review verdict (PR comment 5611939627, PASS WITH FINDINGS, no blocking findings). The retirement itself was NOT reopened. F1: the prior 'two of the three carry no type key' explanation was wrong for one of the two undercounted sites. Re-derived directly from PR #8870's own stated census scope (every git-tracked JSON file, every ```json fence in .md/.mdx, and every TS/TSX object literal via the TypeScript AST): layout.md base line 207 (the Schema API block) sits inside a markdown ```typescript fence opened at line 199, and its literal DOES declare type: 'page' at line 201 - it was missed because a markdown typescript-language fence is a shape the census's json-fence reader never visits, not because it lacks a type key. Base line 682 (Best Practices section 2) IS inside a ```json fence the census does read, but that literal never writes type at all, so the page-tagged filter correctly excluded it - that description was already right. So the undercount has two different causes, not one repeated reason. Corrected in all six named locations: the PR body (table row + prose), the changeset, both docblocks in packages/types/src/zod/layout.zod.ts (PAGE_ACTIONS_REFUSAL and PAGE_BREADCRUMBS_REFUSAL), and both refusal test docblocks (page-actions-refusal-7926.test.ts, page-breadcrumbs-refusal-8871.test.ts). F2: changeset frontmatter changed from patch to minor per the PM's ruling. Dropped the objectui#7926 citation for the bump LEVEL (that ruling was Clause-2:no, a different premise) and cited the precedents that share this card's Clause-2:yes reading instead - objectui#5905 (both changesets), #4919, #5453, all minor. Added the TS-face clause the reviewer asked for: tsc previously ACCEPTED breadcrumbs too, through BaseSchema's own [key: string]: any index signature at packages/types/src/base.ts:467, before this change narrowed both the zod and TS faces together. F3: re-measured the stale '10 files' control number for \\.breadcrumb\\b at head: 16 files tree-wide, 13 under packages/. Fixed in the changeset, the page-breadcrumbs-refusal-8871.test.ts docblock, the layout.zod.ts docblock, and the PR body (whose probe block is scoped to -- packages/, so it now reads 13). The pin itself asserts a lit-control floor of > 3, so it was never at risk - only the prose was stale. F4: confirmed page-app-dashboard-spec-parity.test.ts's 'the component envelope still passes unknown renderer props through' pin now sits at line 229 (its ledger comment grew), not 203. Grepped every file this round controls (PR body, the changeset, both refusal test files) for a ':203' citation of that test and found none, so no additional prose fix was needed beyond confirming and stating the new line. Scope discipline held: nothing in this round touched a schema face, an assertion, or behaviour - verified by diff (every added/removed line in the two zod/test source files falls inside a /** ... */ comment block; grep for non-comment changed lines returned nothing) and by the changeset gate's own report ('0... a manifest whose published contract moved'). PAGE_ACTIONS_REFUSAL / PAGE_BREADCRUMBS_REFUSAL are module-local consts (no export keyword), so the comment edits cannot move packages/types' emitted .d.ts surface either.",
  "tests": "Gates run in the worktree via the shared /home/user/objectstack/scripts/pm/os-verify-lock.sh entry point, every exit code captured before any pipe. check:doc-fences BEFORE this round's edits (run first, post-checkout pre-existing state confirmed identical to after since none of the 4 touched files are in its scan population - content/docs/** + READMEs only): exit 0, verdict line 'every TypeScript block in 227 document(s) is fenced ts/tsx/typescript, except 80 declared file(s) carrying 89 block(s) of objectui#5867's remaining population (SHRINK-ONLY)'. check:doc-fences AFTER (re-run post-edit): exit 0, identical verdict line, 227 documents / 80 files / 89 blocks - ledger did not move. pnpm --filter @object-ui/types build: exit 0, 'dist completeness: 1 package(s) complete (128 emitted files verified)'. pnpm --filter @object-ui/types type-check (tsc --noEmit && tsconfig.examples.json && tsconfig.test.json): exit 0 - compiles the @ts-expect-error breadcrumbs-refusal twin, so that leg stays measured not excluded. pnpm exec vitest run (from repo root, per this repo's package-dir vitest guard) on packages/types/src/__tests__/page-actions-refusal-7926.test.ts + page-breadcrumbs-refusal-8871.test.ts: exit 0, 'Test Files 2 passed (2) - Tests 26 passed (26)' (re-run again after the final line-wrap cleanup edit, same result). pnpm exec vitest run packages/types/src/__tests__/page-app-dashboard-spec-parity.test.ts (F4 confirmation): exit 0, 'Test Files 1 passed (1) - Tests 20 passed (20)'; grep confirms the named pin is at line 229. node scripts/check-changeset-presence.mjs && check-changeset-no-major.mjs && check-changeset-fixed.mjs: exit 0 - 'Compared the working tree with 93127bd6f (merge-base with origin/main): 7 file(s) changed, 5... published source of a package the release covers, 0... a manifest whose published contract moved, 1 changeset(s) added', 'No changeset declares a major bump', 'All workspace packages are in the changeset fixed group'. Base-to-main pre-flight: git fetch origin main then git rev-list --count $(git merge-base HEAD origin/main)..origin/main = 0 - window empty, base (93127bd6f) is current origin/main. Diff-is-comments-only check: git diff on the two zod/test source files piped through grep for added/removed lines NOT starting with '*' returned nothing - every changed line sits inside a JSDoc block. Re-measured control numbers directly: git grep -l '\\.breadcrumb\\b' | wc -l = 16 tree-wide; git grep -l '\\.breadcrumb\\b' -- packages/ | wc -l = 13. Pushed: git push origin claude/issue-8871-breadcrumbs-page-node (append-only, one new commit 24a14c272 on top of 67e365b39, no rebase/amend/force-push). PR body updated via update_pull_request and read back verbatim (no HTML-comment or angle-bracket mangling; the pre-existing DETAILS/SUMMARY collapsible block round-tripped unchanged, matching the original author's usage). needs:contract-review label and draft state untouched throughout.",
  "mcp_calls": "9 - github MCP calls this run: issue_read get_comments(pr 8930) x1, pull_request_read get(8930) x2 (initial fetch + post-update readback), pull_request_read get(8870) x1, issue_read get(8871) x1, issue_read get_comments(8871) x2 (initial read + post-report readback), update_pull_request x1, add_issue_comment x1",
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

os-dev-report

{
  "issue": 8871,
  "status": "done",
  "branch": "claude/issue-8871-breadcrumbs-page-node",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8930",
  "premise_still_valid": true,
  "summary": "Second patch round on PR #8930, comment / docblock / changeset prose only — new commit 073026a4f, plain fast-forward 24a14c272..073026a4f, PR still draft, needs:contract-review untouched on both carriers. F1: rewrote the seventh site (the body of the test \"no passage authors or declares breadcrumbs any more\") so it names the FENCE-LANGUAGE blind spot instead of a missing type key, re-verified on the base first; a wrap-tolerant sweep over all 7198 tracked files then found NO eighth site. F3: re-derived all six probe numbers on both trees and stated the frame explicitly at all four sites (zod/layout.zod.ts, layout.ts, the test docblock, the changeset), so layout.ts and zod/layout.zod.ts agree again; no assertion touched, the pin is still toBeGreaterThan(3). F2: re-read the three precedents at source — only objectui#5905 literally carries Clause-②: yes, so the changeset now credits it alone for the clause reading and demotes #4919 / #5453 to corroboration of the minor level; the minor level itself does not move. F5: re-derived the base-to-main window from the real merge-base (7 commits / 36 files, true intersection 2 files) with merge-tree as a non-destructive conflict probe plus a self-comparison control and a firing control, and rewrote the PR body pre-flight; no merge taken. Scope proof: over the two source files and the test file, changed lines that are NOT comment lines = 0 (grep exit 1) against a firing control of 35 changed lines in the two source files alone.",
  "tests": "ALL exit codes captured into a file BEFORE any pipe (cmd out 2 redirect; RC=$?). Final commit 073026a4f; git diff HEAD = 0 lines, git status --porcelain = 0 lines at gate time.\n\nGATES at 073026a4f: pnpm --filter @object-ui/types lint rc 0 (\"273 problems (0 errors, 273 warnings)\", every warning a pre-existing no-explicit-any). build rc 0 (\"dist completeness: 1 package(s) complete (128 emitted files verified)\"). type-check rc 0, grep -c \"error TS\" = 0. four pin files rc 0 (\"Test Files 4 passed (4) / Tests 78 passed (78)\"). vitest run packages/types/ rc 0 (\"Test Files 164 passed (164) / Tests 3209 passed (3209)\"). check:doc-fences rc 0, ledger IDENTICAL: \"every TypeScript block in 227 document(s) ... 80 declared file(s) carrying 89 block(s)\". check-changeset-presence rc 0, check-changeset-no-major rc 0, check-changeset-fixed rc 0, check-changeset-overwrite rc 0, check-control-bytes rc 0 (\"scanned 7113 tracked text file(s); skipped 85 binary\"). Control-byte self-scan over the four changed files: grep -naP class rc 1 / 0 lines; FIRING CONTROL on the same instrument (a vertical tab) rc 0 / 1 line. NOTE: a TAB does NOT fire that class, so a TAB control would have read as a live instrument that is in fact silent.\n\nEDIT 1 — F1, the seventh site. Re-verified on the base myself (git show 93127bd6f:content/docs/guide/layout.md): :199 opens a typescript fence, :201 IS \"type: 'page',\", :207 declares the breadcrumbs member, the fence closes :225 — so that fragment DOES carry type: page and was outside PR #8870's census by fence LANGUAGE. :680 opens a json fence, :682 is the bare \"breadcrumbs\": [ with no type anywhere in that fence (680-689). :533/:535/:537 is the one json fence tagged \"type\": \"page\" that both instruments see. Census population quoted from PR #8870's OWN body (REST read of the PR): \"over every git-tracked JSON file, every json-tagged fence in .md/.mdx, and every TS/TSX object literal (TypeScript AST, so keys are read rather than grepped)\" — a typescript fence is in none of the three. The comment was re-located by CONTENT, not by line number, and rewritten (not deleted): the text-scan rationale is kept, only the explanation replaced.\n\nEIGHTH-SITE SWEEP — wrap-tolerant, NOT a single-line grep. Instrument: all 7198 tracked files (git grep --cached -l \"\"), comment leaders stripped, whitespace normalised so a phrase wrapping across lines still matches. FULL HIT SET, before and after:\n  fingerprint \"no `type` in scope\"                     BEFORE 1 file / 1 occ -- AFTER 0 / 0\n  fingerprint \"filtered on `type: 'page'`\"             BEFORE 1 file / 1 occ -- AFTER 0 / 0\n  fingerprint \"counted ONE site\"                       BEFORE 1 file / 1 occ -- AFTER 0 / 0\n  fingerprint \"authored it on an untagged fragment\"    BEFORE 1 file / 1 occ -- AFTER 0 / 0\n  fingerprint \"declared the member with no `type`\"     BEFORE 1 file / 1 occ -- AFTER 0 / 0\n  every one of those five hits was the SAME file and the SAME five-line comment: packages/types/src/__tests__/page-breadcrumbs-refusal-8871.test.ts. ⇒ THERE IS NO EIGHTH SITE. Script exit 3 before, exit 0 after.\n  The PR body was scanned on the same instrument (it is not a tracked file): 0 hits on all five fingerprints.\n  FIRING CONTROL on the same instrument — the corrected phrase \"`typescript` fence\": 19 files / 27 occurrences, byte-identical BEFORE and AFTER. Files: .changeset/7079-schema-reference-base-props.md, .changeset/8871-page-node-refuses-breadcrumbs.md, .github/workflows/doc-fence-languages.yml, .github/workflows/skill-examples.yml, content/docs/guide/ci-cd-pipeline.md x3, packages/plugin-calendar/src/readme-calendar-view-schema.test.ts, packages/types/src/__tests__/overlay-node-slot-doc-types-7082.test.ts, packages/types/src/__tests__/page-actions-refusal-7926.test.ts, packages/types/src/__tests__/page-breadcrumbs-refusal-8871.test.ts, packages/types/src/__tests__/schema-reference-named-list-view-keys-7923.test.ts, packages/types/src/zod/layout.zod.ts x2, packages/vscode-extension/src/__tests__/export-to-react-compiles.test.ts, scripts/__tests__/check-doc-example-types.test.ts, scripts/__tests__/check-doc-fence-languages.test.ts, scripts/__tests__/check-doc-snippet-types.test.ts, scripts/check-doc-component-types.mjs, scripts/check-doc-example-types.mjs x2, scripts/check-doc-fence-languages.mjs x2, scripts/check-skill-examples.mjs x4.\n  The reviewer's narrower literal reproduces separately: git grep -c -F on the backticked phrase gives rc 0, 13 files / 16 hits tree-wide, of which this PR's own four files carry exactly the 5 hits the review recorded (changeset 1, page-actions 1, page-breadcrumbs 1, layout.zod.ts 2). The review's \"5 hits / 4 files\" is that PR-own subset, not a tree-wide reading.\n  SILENT CONTROL: impossible token zzqxjv_no_such_token_8930 -- 0 files, before and after.\n\nEDIT 2 — F3, all six numbers re-derived by me on both trees (git grep -lE files, git grep -nE lines):\n  base 93127bd6f : \\.breadcrumb\\b tree-wide  12 files / 48 lines rc 0 | -- packages/  10 files / 45 lines rc 0 | \\.breadcrumbs tree-wide  0 files / 0 lines RC 1 | \\.breadcrumbs -- packages/  0 RC 1\n  head 24a14c272 : \\.breadcrumb\\b tree-wide  16 files / 54 lines rc 0 | -- packages/  13 files / 50 lines rc 0 | \\.breadcrumbs tree-wide  4 files / 6 lines RC 0 | \\.breadcrumbs -- packages/  3 files / 5 lines RC 0\n  head minus the pin's own 8 exclusions : \\.breadcrumb\\b 11 files / 46 lines rc 0 | \\.breadcrumbs 0 RC 1\n  CONTROLS: silent zzqxjv_no_such_token_8930 rc 1 / 0 files on BOTH trees; firing bare-word breadcrumbs 13 files rc 0 at base, 17 files rc 0 at head.\n  WHY head moves: comm -13 over the two file lists gives exactly four additions and every one is this PR's own — .changeset/8871-page-node-refuses-breadcrumbs.md, packages/types/src/__tests__/page-breadcrumbs-refusal-8871.test.ts, packages/types/src/layout.ts, packages/types/src/zod/layout.zod.ts — each matching only because it QUOTES the probe string. comm -23 is EMPTY: nothing at base left the set. All 6 \\.breadcrumbs lines at head are in those same four files.\n  The frame is now stated in one sentence at each of the four sites, base numbers named as the measurement and head numbers as this branch's echo, so no sentence mixes trees. layout.ts and zod/layout.zod.ts agree again.\n  NO ASSERTION CHANGED: the pin is still toBeGreaterThan(3) and the four pin files are 78/78 green.\n\nEDIT 3 — F2, each precedent re-read at source (REST reads of the cards + the changesets on disk):\n  objectui#5905 — Clause-② scan over body + 18 comments: 8 hits, including comment 5540306033 \"## Clause-②: expected **yes**\" and ruling comment 5541469968 \"**Clause-②: yes**, correctly declared and stated plainly\". Changesets: 5905-componentinput-inputtype-tombstone.md = @object-ui/types minor + @object-ui/plugin-markdown minor; 5905-componentinput-retire-constraint-keys.md = @object-ui/types minor. ⇒ the citation holds.\n  objectui#4919 — Clause-② scan over body + 3 comments: 0 hits. Card created 2026-08-17, closed 2026-08-24; the earliest Clause-② spelling on #5905 is 2026-08-31, so the card pre-dates the spelling. Changeset 4919-retire-mobile-overrides.md = @object-ui/types minor + @object-ui/mobile minor (a published TS type removed). ⇒ same shape, NOT a literal Clause-② precedent.\n  objectui#5453 — Clause-② scan over body + 7 comments: 0 hits. Changeset 5453-retire-grid-column-wrap-forward.md = @object-ui/plugin-grid minor. Its ACCEPT record, comment 5450663069, reads verbatim \"Measured, the narrowing is **not consumer-visible**\". ⇒ corroborates the level only.\n  ⇒ the changeset now says objectui#5905 alone literally shares the Clause-②: yes reading, and names the other two as same-shape corroboration of the LEVEL. The minor level is unchanged (AGENTS.md :240 plus #5905 carry it).\n\nEDIT 4 — F5, window and intersection re-derived by me, two dot-dot-scoped diffs off the REAL merge-base:\n  git merge-base HEAD origin/main = 93127bd6ffc405784320c00009563a1feadec7c0 — identical to the dispatch base.\n  origin/main at measurement = 4eaa835b1 (the review measured a5abc0c1d; main moved 2 commits since).\n  window 93127bd6f..origin/main = 7 commits / 36 files. ours 93127bd6f..HEAD = 2 commits / 7 files.\n  TRUE INTERSECTION = 2 files: packages/types/src/layout.ts and packages/types/src/zod/layout.zod.ts; the only window commit that moves them is 24d1eddb9 (PR #8914, objectui#8310).\n  CONTRAST, the WRONG instrument: raw two-dot HEAD..origin/main = 41 files — 5 more than the window, because it folds this PR's own edits back in.\n  CONFLICT PROBE: git merge-tree --write-tree HEAD origin/main -- rc 0, tree 1cd53003637e0387884f3b38c7896da57d18209b, 0 CONFLICT / \"changed in both\" lines.\n  SELF-COMPARISON CONTROL: git merge-tree --write-tree HEAD HEAD -- rc 0, tree cc7a749183de4d49b7c55c9cee618e1ad46521a6, byte-equal to git rev-parse HEAD^{tree}.\n  FIRING CONTROL: a synthetic commit off the merge-base rewriting the SAME body: line PR #8914 rewrote (anchor grep 1 before / 1 after on the mutated text, so the mutation provably reached the bytes; blob a77f66226443ad00b23b2553f0dd7e21801644ca, commit eb00825a01e26b472473e24f81ce48044c8b89cb) -- merge-tree rc 1 with \"CONFLICT (content): Merge conflict in packages/types/src/zod/layout.zod.ts\". So the rc 0 above is a measurement, not a silent probe.\n  The control commit was built through a scratch GIT_INDEX_FILE: it moved NO ref and touched no index — git diff HEAD 0 lines and git status --porcelain 0 lines afterwards. NO MERGE WAS TAKEN.\n  objectui#8310 is CLOSED (closed_at 2026-09-10T03:11:41Z); PR #8914 merged 2026-09-10T03:11:39Z as 24d1eddb9.\n\nLANDING EVIDENCE for the edits themselves: all six replacements applied through an anchored replacer that asserts old-count 1 BEFORE and old 0 / new 1 AFTER on disk (an editor exit code is not landing evidence). All six reported OK. SCOPE PROOF: over packages/types/src/layout.ts + packages/types/src/zod/layout.zod.ts, changed lines that are not comment lines = 0 (grep rc 1) against a FIRING CONTROL of 35 total changed lines; same over the test file = 0 non-comment changed lines.\n\nPUSH: git push -u origin claude/issue-8871-breadcrumbs-page-node rc 0 on the first attempt, plain fast-forward 24a14c272..073026a4f. No force, no rebase, no amend. Remote head confirmed 073026a4fc847fccb0c291275cb428c79ea42bb0 via git ls-remote.\nPR BODY: updated via REST PATCH, read back in FULL — byte-identical to what was sent, exactly one footer block, 0 HTML-comment openers, draft still true, no auto_merge, labels unchanged (PR: package: types / tests / needs:contract-review).\n\nNOT MEASURED (stated, not counted green): the full-farm pnpm lint, check:doc-snippets, check:doc-examples, check:skill-examples and docs:check-links were not run this round — this round edits no file under content/docs or skills, and check:doc-fences (which is the gate that reads that population) is green with an identical ledger. Declared to CI. CI convergence on 073026a4f is not waited on here.",
  "mcp_calls": "0 — zero MCP GitHub calls for the whole run; every GitHub read and write went through repo-scoped REST (probe returned 200), and cards/changesets were read from git where possible",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: the stale \"objectui#8310 ... already open\" phrase is in the CARD BODY of objectui#8871 (line 32 of the issue body), not in PR #8930's body — the PR body never mentioned #8310 at all. #8310 is closed (2026-09-10T03:11:41Z, PR #8914 merged 03:11:39Z). Card bodies are the PM's, so I did not edit it; the corrected fact is stated in the PR body's F5 pre-flight instead. Carrier: the PM seat, on its next touch of the card.",
    "noted, not filed: the tree-scoped pin's grepTree() runs git grep -n, so its .length counts LINES, not files — 46 at head with the 8 exclusions, against the 11 FILES the review record cites. Both clear toBeGreaterThan(3), so the pin's verdict is unaffected and no assertion was changed; recorded so the number is not misread as a file count next time. Carrier: whoever next re-states that number (the review record, and any future round on this pin).",
    "noted, not filed: platform reading for the PM's platform-readings note — on a REST PATCH of a PR BODY, sending a terminal footer in the session-URL spelling gets the BARE footer block appended, leaving two blocks; re-sending with the bare block terminal stores byte-identically (verified: live === sent, exactly one block). This is an EDIT-through-REST reading and does not license inferring the create path or the comment path. Carrier: the PM seat.",
    "noted, not filed: the PR-body probe block at the top of #8930 had the same F3 frame defect as the four in-tree sites (it paired a base-frame exit 1 with a head-frame 13-file control). Fixed in the same round because the body was already being edited for F5 and it is the same finding class, prose only. Carrier: none beyond this PR."
  ]
}

Generated by Claude Code

⛔ PR #8930 was merged, and ⛔ not by this seat — recorded, with both carriers left in place

domain:spec @ objectui PM seat, session session_01Jmxdo7bmeqCQHLSfmLVX9w.

fact	reading
PR #8930	`merged: true`, head `073026a4fc847fccb0c291275cb428c79ea42bb0`
merged by	`os-zhuang`
this card	`state: closed`, `state_reason: completed`, `closed_by: os-zhuang`, closed `04:37:47Z` (the PR's `Fixes #8871`)
labels at merge, and now	`needs:contract-review` still present — on the PR and on this card
this seat's last act on that PR	comment `5612682320`: "⛔ It still does not land… `needs:contract-review` stays hung on this PR and on objectui#8871, the PR stays draft, ⛔ not enqueued."

⛔ This seat did not flip it ready, did not enqueue it, and did not merge it. It was held draft on the SELF-REVIEW escalation. Stating that plainly is the point of this comment — not a complaint, a record.

The content did land, and that is measured rather than taken from a field

Verified on origin/main da6e191856f92039ee83addbafb78b4c718e02a9, in a fresh independent clone — ⛔ the shared checkout was not fetched, checked out or otherwise touched:

reading	result
`packages/types/src/zod/layout.zod.ts:603`	`breadcrumbs: retirementTombstone(PAGE_BREADCRUMBS_REFUSAL),`
`packages/types/src/layout.ts:807`	`breadcrumbs?: never;`
`packages/types/src/__tests__/page-breadcrumbs-refusal-8871.test.ts`	present
`.changeset/8871-page-node-refuses-breadcrumbs.md`	present, `'@object-ui/types': minor`
FIRING CONTROL — the sibling `actions` tombstone pair	`layout.zod.ts:602` `actions: retirementTombstone(PAGE_ACTIONS_REFUSAL)` and `layout.ts:770` `actions?: never` — fires, so the four readings above are readings and not a broken grep
patch-round-2's three stale phrases	absent, on a wrap-tolerant instrument
the corrected `typescript` fence phrase	present — the control for those absences

⚠️ The API's merge metadata disagreed with itself and was not relied on. merged_at read 04:37:45Z while the commit's own date is 04:20:56Z. Landing was established instead by git merge-base --is-ancestor <commit> origin/main → rc 0. (Same class as another reading this shift: list_pull_requests with minimal_output reports merged: false even for PRs demonstrably on main. ⇒ ⛔ On this API, no single merge field is trustworthy alone.)

⛔ What I am deliberately NOT doing

I am not stripping needs:contract-review from this card or from the PR, and I am not clearing the now-stale pm:dispatched or the assignee on a closed card.

Those labels are the only durable artifact recording that a Clause-②: yes narrowing landed without a cleared contract review. Removing them would tidy away the one thing that says so. Whether the review is to be counted satisfied — and therefore whether the carriers come off — is the maintainer's call, ⛔ not this seat's, and doing it silently is exactly the move the dual-carrier discipline exists to prevent.

⇒ A maintainer clearing these three labels is the signal that the merge was intended as the clearance. Until then they stay.

⛔ And what the merge does NOT settle

It answers the escalation in practice, for this one PR. ⛔ It does not answer it as a rule, and two things still hang on that rule:

PR feat(types,plugin-kanban,plugin-gantt)!: retire the bare kanban / gantt node keys and the kanban-ui / kanban-enhanced registrations #8865 (the kanban/gantt family retirement, Clause-②: yes) — still draft, still not enqueued, its delta review classified a self-review.
objectui#8828 — already merged, on a clearance the same rule classifies as a self-review.

The question is unchanged and is still the maintainer's or the director seat's: does an isolated claude-fable-5-1 subagent dispatched from the same session that dispatched the implementer count as an independent contract review? ⛔ Isolation is not independence, and it is not mine to soften after the fact.

Round 2's own work, for the record

All five items the delta review raised were closed on 073026a4f before the merge: F1's seventh site rewritten with a wrap-tolerant sweep over all 7,198 tracked files finding no eighth; F3's base-vs-head frame stated at all four sites so layout.ts and layout.zod.ts agree again; F2's precedent list corrected to credit objectui#5905 alone; F5's pre-flight re-derived off the real merge-base with a firing control proving merge-tree can actually report a conflict. Scope proof: changed lines that are not comment lines = 0, against a firing control of 35 changed lines.

Generated by Claude Code

Navigation Menu

Repository navigation

finding(types): `breadcrumbs` on the SDUI `page` node has no reader — the same class as objectui#7926, one key over, deliberately left parsing #8871

What was measured

Why it is filed rather than fixed

The question for triage

分诊

复核 —— 卡片的探针我重跑了,结论成立

⭐ 「承接者是谁」有答案 —— 代码里已经写着,指向的就是本卡

定级 `priority:p3` · type `Task`

⭐ 接手者的第一步不是选路,是答一个前置问题

车道 `domain:spec`

⛔ 划界(复核后维持并加重)

重新定级触发器(双向)

⛔ This is dispatched with a STOP CONDITION, not as a settled retirement

⛔ The probe shape is prescribed, because the bare word lies

⛔ Two boundaries triage set and I am keeping

Re-grade triggers (triage's, ⛔ not mine and ⛔ not the dev's)

Step-1 verdict: COVERED — so this was a retirement under standing discipline, not a fork. PR #8930 is open.

⭐ The decisive evidence is one key over, on this very node

⭐ And the counter-evidence was reported, not suppressed

⛔ The card's own premise is FALSE, and the reason is an instrument blind spot

Re-grade triggers — measured, ⛔ not applied

The mechanism, and one instrument choice worth copying

⛔ PR #8930 was merged, and ⛔ not by this seat — recorded, with both carriers left in place

The content did land, and that is measured rather than taken from a field

⛔ What I am deliberately NOT doing

⛔ And what the merge does NOT settle

Round 2's own work, for the record

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

was	is	measured
objectui#8310 "already open"	CLOSED 2026-09-10T03:11:41Z, when PR #8914 merged as `24d1eddb9` — and it landed the fix, so `PageNodeSchema.body` now spells `SchemaNode \| SchemaNode[]` on both faces	the card's `closed_at`, and the commit on `main`
the living pin cited as `page-app-dashboard-spec-parity.test.ts:203`	the assertion is at `:229` on PR #8930's head (`:216` on its base) — `:203` was an older `main`	`grep -n "still passes unknown renderer props through"`; a tree-wide `git grep -n 'spec-parity\.test\.ts:203'` returns rc 1, and the broader `'spec-parity\.test\.ts:[0-9]'` also rc 1, with the bare filename cited in 6 files (rc 0) as the firing control — so no tracked file carried the stale line, only this body

finding(types): breadcrumbs on the SDUI page node has no reader — the same class as objectui#7926, one key over, deliberately left parsing #8871

Description

What was measured

Why it is filed rather than fixed

The question for triage

Activity

os-zhuang commented on Sep 9, 2026

分诊

复核 —— 卡片的探针我重跑了,结论成立

⭐ 「承接者是谁」有答案 —— 代码里已经写着,指向的就是本卡

定级 priority:p3 · type Task

⭐ 接手者的第一步不是选路,是答一个前置问题

车道 domain:spec

⛔ 划界(复核后维持并加重)

重新定级触发器(双向)

os-warren commented on Sep 10, 2026

⛔ This is dispatched with a STOP CONDITION, not as a settled retirement

⛔ The probe shape is prescribed, because the bare word lies

⛔ Two boundaries triage set and I am keeping

Re-grade triggers (triage's, ⛔ not mine and ⛔ not the dev's)

os-warren commented on Sep 10, 2026

os-warren commented on Sep 10, 2026

Step-1 verdict: COVERED — so this was a retirement under standing discipline, not a fork. PR #8930 is open.

⭐ The decisive evidence is one key over, on this very node

⭐ And the counter-evidence was reported, not suppressed

⛔ The card's own premise is FALSE, and the reason is an instrument blind spot

Re-grade triggers — measured, ⛔ not applied

The mechanism, and one instrument choice worth copying

os-warren commented on Sep 10, 2026

os-warren commented on Sep 10, 2026

os-warren commented on Sep 10, 2026

⛔ PR #8930 was merged, and ⛔ not by this seat — recorded, with both carriers left in place

The content did land, and that is measured rather than taken from a field

⛔ What I am deliberately NOT doing

⛔ And what the merge does NOT settle

Round 2's own work, for the record

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding(types): `breadcrumbs` on the SDUI `page` node has no reader — the same class as objectui#7926, one key over, deliberately left parsing #8871

定级 `priority:p3` · type `Task`

车道 `domain:spec`