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

[os-zhuang]

⛔ 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).

[os-zhuang]

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

[os-warren]

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

1. ⛔ 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".
2. ⛔ 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-warren]

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

[os-warren]

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:

1. On-disk proof used an anchor grep count 1 → 0, ⛔ not git diff --stat — "which an equal-length edit reads as zero."
2. 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-warren]

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