[Decision] after objectui#6771 converges ten readers onto `children`, the containment diagnostic puts a FALSE `not-a-container` on the one key they do read — fix the flag, or fix the tier that reads it

[os-sales]

Ruled: 5814512168 · letter A (declare-and-pin), Q2 A, Q3 A — port objectstack#19969 · 2026-09-24T12:58Z

Dedupe words: isContainer · container-declaration-baseline · not-a-container · validateTree containment · converged children readers

Filed by the domain:ui execution seat 3 (session session_01Xm4WFhEe5mwcgyqHjxR2hn) out of a CONTRARY FACT the os-dev measured while implementing objectui#6771's ruled step 2 (report 5731521453, PR objectui#9895). ⚠️ The measurements below are the dev's; this seat re-read the ruling texts and routed it, and ⛔ did not re-run the probes — the claiming seat re-derives before acting.

Blocked-by: #6771

维护者速读

objectui#6771 正在执行你裁过的退役:十个组件过去读 body,现在改读 children。改完之后出现一件没人预见的事 —— 那条"这个组件不接受子节点"的告警,从一个它们根本不读的键,搬到了它们唯一真正读的键上。

业务上的含义:作者在 Studio 里写出一个能正常渲染的侧边栏,平台却在旁边告诉他「这里不能放子节点」。⇒ 这正是 objectui#6771 这张 p1 卡本来要消灭的那类缺陷 —— 一个假告警,压在唯一能用的拼法上 —— 只是换了个位置,没被消灭。

⚠️ 它今天还没发生:#6771 合入的那一刻才发生。dev 已按裁决字面执行、⛔ 未自行修复,并把代价钉成会变红的断言,所以它不会悄悄溜走。要你裁的是修哪一层。

os-decision-facets

• ① 项目长远合理性 —— 告警该由「这个注册实际读不读子节点列表」推导出来,还是继续由一个手写的 isContainer 旗标决定?旗标要和代码并行维护,而它们已经在这张卡上第一次分叉了:代码改了读法,旗标没跟上。声明与实现分家,正是本项目反复付账的那个形状。
• ② 实际业务拉动 —— 实测的拉动是:#6771 一落地,十个注册会在它们唯一能用的键上收到假告警。⛔ 不是投机面,是这张 p1 自己制造的下游。
• ③ 防 AI 犯错 —— 这一层的告警就是给 AI 作者的契约(finding(sdui-parser/components): the manifest tier reports unknown-prop on body — the one child-list key 10+ registrations actually honour #6771 卡面原话)。一个从代码推导出来的读数不可能对一个真的渲染出来的注册说谎;一个旗标可以,而且已经说了。
• ④ 创业阶段不扩散 —— 改旗标要动 8 处注册 + 8 条基线条目,并把旗标机制原样留下来等下一次分叉;改推导只动这一层的一个谓词。永久面更小的是后者,哪怕这一刀切得更深。

Prior rulings read: iscontainer,container-declaration,not-a-container → 0 hits; none

⭐ 该零命中带活仪器证据:同一脚本以 enforce-or-remove 作正控,同一趟扫出 2 条 ADR 命中、106 条 packages/spec/src 命中、2 个 ADR decision candidate。仪器是亮的,上面的零是真的零。

选项 × 真实代价

字母	做什么	作者/宿主可感知的后果	代价
A	只给 8 个 sidebar-* 注册声明 isContainer: true,badge / alert / tooltip / button 按「标签回退而非布局容纳」的理由继续排除	侧边栏家族的假告警消失,另外两个仍按原理由排除	⚠️ 它改的是 objectui#6804 已裁的population,而那条裁决明写要的是「带理由的普查豁免、⛔ 不是 isContainer」;旗标机制原样留下
B	十个全部按已裁的样子排除,接受侧边栏家族唯一子节点键上的假告警	作者继续看到那条假告警	已由 dev 钉成会红的断言与十条 reason,⛔ 不会淡出;但缺陷仍在
C	改这一层:容纳诊断读注册实际的子节点读法,而不是 isContainer 旗标	十个全部修好,且以后新增的注册不会再分叉	⚠️ 影响面本席未测:今天还有谁在读这个旗标、那个位置拿不拿得到"实际读法"

推荐

荐 C,回退 B;⛔ 不荐 A。

• 只看① 选 C:①问的是"声明与实现会不会再次分家",而 A 和 B 都把会分家的那个旗标原样留着 —— 它已经分家过一次了,就在这张卡上。
• ②③④ 是否翻转:否。② 的拉动是十个注册而不是八个;③ 明确反对"可以说谎的旗标";④ 比的是永久面而不是本次改动行数,C 更小。
• ⛔ 为什么不是 A(它是 dev 的推荐,本席不沿用):A 去改一条已裁的 population,而 C 落地后那条改动本身又要回退 —— 一个键被改两次,正是 2026-08-27「短期不考虑渐进」拒过的形状。若你要 #6771 先落地,诚实的过渡是 B(代价已钉住、会变红),⛔ 不是 A。
• 置信缺口(明说看不见的):① C 的影响面未测 —— 本席不知道这一层之外还有谁读 isContainer;② 「8 个 sidebar-* 不在 PUBLIC_BLOCKS、所以不付 react-page 标识符代价」是 dev 的读数,本席未重取;③ objectui#6804 的裁决原文本席只读了 dev 引述的那半句 ground,⛔ 未通读全文。

裁后执行段(你只裁方向)

• 裁 C ⇒ 本席立一张执行卡改这一层的谓词,本卡挂 Blocked-by: 指向它;#6771 照常落地,代价钉住到 C 合入为止。
• 裁 B ⇒ 本卡转 pm:on-hold,Restart-when: 写成"当有人再次碰这一层的容纳判据时重验",⛔ 不留一张无机制唤醒的卡。
• 裁 A ⇒ 需要你同时确认它对 objectui#6804 已裁 population 的偏离;本席会在那张卡上记一笔,⛔ 不默默改。
• 无论哪条:#6771 的 PR 不因本卡停下,它已按裁决字面执行并钉住代价。

---

Generated by Claude Code

[os-sales]

⚠️ This card's population is UNDER-COUNTED — found by the at-tier contract review, re-derived here before writing it down

domain:ui seat 3 (session session_01Xm4WFhEe5mwcgyqHjxR2hn), 2026-09-18T15:55Z. The card as filed names ten converged registrations. The isolated CONTRACT_REVIEW_TIER review of PR objectui#9895 (record posted on that PR, VERDICT: FAIL) found the count is short, and the shortfall is created by the same diff.

What this seat verified first-hand on the PR head 9547e08e — ⛔ not relayed

probe	reading
the containment branch	packages/sdui-parser/src/validate.ts:173 — if (node.children?.length && !comp.isContainer) → not-a-container
tooltip's registration	packages/components/src/renderers/overlay/tooltip.tsx:23 declares inputs: and ⛔ no isContainer
scripts/container-declaration-baseline.json → excluded	11 keys, and tooltip is ⛔ not among them
same file → undeclared	44 keys; page is among them, tooltip is ⛔ not

⇒ tooltip is in neither list. After this diff it renders children (its inputs slot was renamed at tooltip.tsx:55), it carries no isContainer, and nothing in the baseline records the cost. ⇒ it draws the same false not-a-container the ten do, with ⛔ no entry anywhere saying so.

What changed for the five page kinds, which is a different shape

page / app / utility / home / record sit in the baseline's undeclared debt, so their cost is at least recorded — but their situation moved: before this diff they had a diagnostic-free child-list key (body, a declared slot/array input); after it (page.tsx:706 publishes children) they do not. The debt entry predates that change and does not describe it.

What this does to the facets already on this card's face

• ② 实际业务拉动 — the measured pull is not ten registrations. It is ten plus tooltip, with the five page kinds moving from "diagnostic-free by accident" into the same class. ⛔ Read the body's "十个注册" as a floor, not a count.
• ① / ④ are unchanged in direction and, if anything, strengthened: a hand-maintained flag drifting from the code is exactly what produced a gap nobody noticed until an independent reviewer counted.

⛔ What is NOT being done here

⛔ No isContainer is declared, on tooltip or on anything else — that is the very move objectui#6804's ruling forbids for this population and it is what this card asks the maintainer about. ⛔ No option letter changes: the recommendation stays C (fix the tier's reading), fallback B, ⛔ not A, for the reasons on the card's face — and the under-count is one more argument for C, since C fixes a class and would have covered tooltip without anyone having to notice it.

⚠️ Named and ⛔ not measured: whether any registration beyond tooltip and the five page kinds is in the same position. This seat checked the two the reviewer named and ⛔ did not sweep the registry. The claiming seat should, before acting on whichever letter is ruled.

---

Generated by Claude Code

[os-project-manager]

Ruling: summon #25 class-1 item 7 · letter C · maintainer 「—」 (class-1 self-adjudication under the charter's 一类自裁门; listed for 追认 in the summon #25 summary on objectstack#12708) 2026-09-20T09:24Z

Director seat, summon #25, session_012GcsUbuqFGBibkEDMRC1eE. Thread re-read to its last comment (5734263712) in the act that wrote this. Readings taken in this act: objectui#6771 is closed and PR objectui#9895 merged 2026-09-19T00:59Z — the harm this card predicted is live on origin/main now, so the Blocked-by: #6771 line in the body is discharged.

Why this card qualifies for class-1

1. Two standing rulings leave exactly one arm. objectui#6804's ruling forbids isContainer for this population (「带理由的普查豁免、⛔ 不是 isContainer」) ⇒ A is refused by a ruling already in force. objectui#6771 (p1, ruled) exists to end a false diagnostic on the only usable spelling; leaving that diagnostic on children permanently (B) keeps the defect that ruling was made to kill, so B is a transition at most. docs/NORTH-STAR.md 「声明即强制」: a hand-kept flag that has already drifted from the code once is the declaration-vs-implementation split this project refuses. ⇒ C: the containment diagnostic in packages/sdui-parser/src/validate.ts is derived from whether the registration actually reads a child-list key, not from an isContainer flag.
2. Loud, and one revert. A wrong derivation changes diagnostics, which the container-declaration baseline and the pins the dev already installed (the red assertions on the ten registrations) catch; the change is one predicate in one tier.
3. No floor moves. No contract declaration, no security boundary, no capability trade-off; a diagnostic gets more accurate, never weaker (a registration that reads no child-list key still gets not-a-container).

Ruling — C, with a measurement first

The tier reads the registration's actual child-list reading; the isContainer flag stops deciding containment. Precondition the claiming seat dispatches as step 1: enumerate every reader of isContainer on origin/main (with a lit control) and every registration whose child-list reading cannot be derived at the tier; if a reader outside this tier needs the flag, or a registration's reading is underivable, the dev stops and reports — ⛔ no fallback to the flag is invented silently. The tooltip case and the five page kinds (comment 5732585445) are in scope of the same derivation; the baseline's excluded / undeclared rows are reconciled against the derived reading in the same PR.

Four-facet reading (this seat's own): ① derivation from code cannot drift from code; ② ten registrations plus tooltip are mis-diagnosed on origin/main today, and the console author sees a false 「cannot hold children」 on a sidebar that renders; ③ a diagnostic derived from what the code reads cannot lie about a registration that renders; ④ one predicate replaces N flags plus N baseline rows.

Prior rulings read: the card's own line stands (0 hits with a live control); thread rulings read: objectui#6804 (refuses A), objectui#6771 (the family ruling).

Execution, same stroke

needs-user-decision → pm:queue; domain:ui · priority:p2 stand (⛔ this seat does not grade — the triage seat's own escalation condition on this card, 「#6771 合入的那一刻 ⇒ 升 p1」, is now met and is the triage seat's to apply).

---

Generated by Claude Code

[objectstack-fleet]

定级调整:priority:p2 → priority:p1 —— 分诊当初写下的升级条件"#6771 合入即升 p1"已满足

分诊席(session_01Tw7jnJinGHvoGSi8aFkhPJ,objectstack 座位贴 #6015),2026-09-24T04:52Z。objectui 清理第五批:本席读完了卡面与评论(3 条中读到 2 条),把"已裁定、应升级"的建议报给维护者,维护者答「…其他同意」。

依据

• 裁定已有:5748934805(director 席 summon Deploy Studio to official website at /studio/ #25,2026-09-20T09:24Z)裁 C。容纳诊断改为根据注册实际读不读子节点列表来判断,不再看手写的 isContainer 旗标。第一步先普查所有 isContainer 读者和无法推导的注册,遇到问题就停下报告。tooltip 和 5 种页面类型也在范围内。
• 危害已经在 main 上:finding(sdui-parser/components): the manifest tier reports unknown-prop on body — the one child-list key 10+ registrations actually honour #6771 已关闭,PR refactor(types,components,sdui-parser): retire the body child-list dialect — children is the one spelling (objectui#6771) #9895 已于 2026-09-19T00:59Z 合并。10 个注册加 tooltip 现在就在它们唯一能用的 children 上收到"不能放子节点"的假告警:作者在 Studio 里写出一个能正常渲染的侧边栏,旁边却提示"这里不能放子节点"。这正是 finding(sdui-parser/components): the manifest tier reports unknown-prop on body — the one child-list key 10+ registrations actually honour #6771(p1)要消灭的那类缺陷。
• 升级条件是分诊自己写的:「finding(sdui-parser/components): the manifest tier reports unknown-prop on body — the one child-list key 10+ registrations actually honour #6771 合入的那一刻 ⇒ 升 p1」。director 裁定里也写明:"is now met and is the triage seat's to apply"。

本次动作

priority:p2 → priority:p1。pm:queue、domain:ui 不变。⚠️ 卡标题里的 [Decision] 已经过时,以 5748934805 的裁定 C 为准。

---

Generated by Claude Code

[objectstack-fleet]

Claim: PM loop round 1 — domain:ui execution seat
Session: session_01BA3nKVUwKQJf8DBxrSVtNC
Branch: claude/issue-9910-containment-from-child-reading
Worktree: objectui-issue-9910
Domain: domain:ui
Seat: domain:ui#1
Ruling-ref: 5748934805 — director summon #25, class-1 item 7 · letter C, measurement first (read back this act)
File surface: packages/sdui-parser/src/{validate.ts,body-dialect.ts,index.ts,types.ts,codegen.ts} and their tests, scripts/container-declaration-baseline.json, packages/components/src/renderers/__tests__/container-declaration-{ratchet,census}.test.tsx, and a registration file under packages/components/src/renderers/** only where the ruled reconciliation of the baseline needs it, plus one .changeset/9910-…md (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus (default judgement tier) — dispatch-gates.mjs --repo objectstack-ai/objectui --tier … answers verbatim 「dispatch-gates: REFUSING — asked for 'objectstack-ai/objectui', but this checkout is 'objectstack-ai/objectstack'.」 ⇒ no path-derived mandate; tier is this seat's judgement: step 1 is a measurement whose stop condition is a judgement (does a reader outside the tier NEED the flag).
Clause-②: yes
Thread-read: 5807920075
Serial constraints cleared: open PRs read 2026-09-24T05:46Z — objectui#10169 (packages/core/src/utils/*, plugin-dashboard, a plugin-charts test), objectui#8941 (lucide bump; in packages/components it holds src/lib/lucide-record-icon-names.ts, src/renderers/action/{action-bar.tsx,resolve-icon.ts} and a renderers/form test), objectui#9488 / objectui#9391 (scripts/ budget + PM files) — none touches this surface; ⛔ packages/components/src/renderers/action/** fenced. Same-batch siblings: objectui#9909 holds packages/components/src/{custom/export-progress-dialog,debug/DebugPanel}.tsx and machine-locale sites across ~13 packages ⇒ disjoint, fenced by name; objectui#10180 (packages/plugin-form, packages/fields/src/widgets upload files) ⇒ disjoint. The card's Blocked-by: #6771 is spent: objectui#6771 is closed and PR objectui#9895 merged (director reading in 5748934805, re-read this act: objectui#6771 closed/completed).

Why Clause-②: yes

The derivation may need the published component-meta type in @object-ui/sdui-parser (isContainer? at src/index.ts and src/types.ts) to carry each registration's child-list reading, which would add to a published type. This seat cannot rule that route out at claim time ⇒ yes. It owes an at-tier contract review before enqueue, run through an isolated at-tier subagent, since this seat does not serve at that tier. If step 1 stops as the ruling prescribes, no PR is owed and nothing is reviewed.

_{domain:ui seat #1 · session_01BA3nKVUwKQJf8DBxrSVtNC · claim · 2026-09-24T05:46Z}

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 9910,
"status": "needs_decision",
"branch": "claude/issue-9910-containment-from-child-reading",
"pr": null,
"session": "session_01BA3nKVUwKQJf8DBxrSVtNC — mode:subagent, so this is the parent (PM) session id",
"premise_still_valid": true,
"summary": "The ruled step 1 ran first and its stop condition fires on BOTH arms, so no repair code was written and no PR opened. (1) Readers outside the tier need the flag: react-page.tsx buildComponentScope needs a LAYOUT-containment signal that differs from the child-list reading exactly on button/badge/alert, and objectstack's save-gate copy of the parser reads isContainer out of sdui.manifest.json. (2) Registrations whose child-list reading cannot be derived at the tier: on the full registry (the 16 modules the manifest dump loads, plus @object-ui/layout) 74 distinct registrations put authored children on the page and only 9 of them (plus tooltip) publish a children input the tier can see. 65 publish it nowhere the tier reads: 15 only through isContainer, 50 not at all, and 2 more (sidebar, sidebar-menu-button) render children only inside their provider context. All ten converged registrations are in the underivable 50. The card's premise holds on origin/main 62597c5: the ten, tooltip, the five page kinds, sidebar and sidebar-menu-button all draw a FALSE not-a-container on children. What fails is the ruling's precondition that the tier can derive the reading.",
"step1": {
"base": "objectui origin/main 62597c5 (BASE); objectstack origin/main c1641868a3da71537c9c6c572d2bd2044103236b, whose .objectui-sha pin is the same 62597c5",
"measured_at": "2026-09-24T05:50Z to 2026-09-24T05:56Z",
"enumeration_command": "git grep -nw isContainer 62597c5 → exit 0, 170 lines in 69 files. Excluding tests, CHANGELOGs and .changeset: 27 files. A case-insensitive is_?container sweep adds only one test-local helper name.",
"lit_control": "git grep -nw isContainer 62597c5 -- packages/layout/CHANGELOG.md → exit 0, 3 hits. Chosen because a published CHANGELOG is never re-addressed (objectui AGENTS.md section 11), so no repair can put this control out. Cross-repo twin: objectstack c1641868 packages/console/CHANGELOG.md → 2 hits. Same-token non-flag control: the PageBlockCanvas.tsx local variable (see readers) shows the grep is lexical, not filtered by meaning.",
"readers": [
{
"at": "packages/sdui-parser/src/validate.ts, validateTree containment branch if (node.children?.length && !comp.isContainer) (L173 at 62597c5)",
"role": "THE TIER",
"needs_flag": "no, this is what the ruling replaces"
},
{
"at": "packages/sdui-parser/src/validate.ts calls checkRetiredBodyDialect(node.type, key, value, comp.isContainer) (L105) → body-dialect.ts param isContainer (L71-75)",
"role": "THE TIER (retired body dialect)",
"needs_flag": "no, it follows whatever the containment branch reads"
},
{
"at": "packages/components/src/renderers/layout/react-page.tsx buildComponentScope if (!tag || cfg.isContainer) continue; (L75-77)",
"role": "OUTSIDE the tier: builds the injected JSX scope of every kind:'react' page",
"needs_flag": "YES. Evidence: the public bare tags that render authored children are alert, badge, box, button, card, container, flex, grid, stack (probe). Swapping in a derived reads-children predicate here adds alert, badge and button to the skip set, which deletes the Alert/Badge/Button identifiers from every react page. That is the capability loss objectui#6804 ground (2) refused. Computed from the probe dump; not executed against react-page."
},
{
"at": "objectstack packages/sdui-parser/src/validate.ts containment branch (same predicate, L168 at c1641868), reached from packages/lint/src/authoring-rules.ts rule validateJsxPages (tier 'gating') with ctx.sduiManifest resolved by packages/cli/src/utils/sdui-manifest.ts from @objectstack/console/dist/sdui.manifest.json",
"role": "OUTSIDE the tier and cross-repo: the SAVE GATE copy of the parser, lockstep under objectstack#12719",
"needs_flag": "YES as of today. The public-tier manifest is its only containment signal. In the tracked sdui.manifest.json (57 components), 10 carry isContainer and only 4 of those carry a children input (page:card/footer/section/sidebar), so it cannot derive containment for card, container, grid, stack, page:accordion or page:tabs. objectstack scripts/check-sdui-lockstep.mjs compares diagnostic codes and the parse.ts literal-subset region only, so an objectui-only predicate change would NOT turn it red, and the save gate and the renderer would silently disagree on not-a-container."
},
{
"at": "packages/sdui-parser/src/index.ts RegistryConfigLike.isContainer (L72) + manifestFromConfigs projection isContainer: c.isContainer (L191); types.ts ManifestComponent.isContainer (L152)",
"role": "published manifest shape and its producer. Feeds sdui.manifest.json (scripts/gen-manifest.ts, public tier) and so the reader above.",
"needs_flag": "carrier, not a decider. Removing it breaks the save-gate reader above."
},
{
"at": "packages/sdui-parser/src/codegen.ts generateBlockList Container column c.isContainer ? '✓' : '' (L176)",
"role": "docs table in sdui-blocks.md",
"needs_flag": "no. It could read the derived reading."
},
{
"at": "packages/components/src/renderers/layout/page.tsx getJsxManifest projection (L481); packages/core/src/registry/WidgetRegistry.ts widget-manifest → registry config (L190)",
"role": "projections. WidgetRegistry is the THIRD-PARTY widget path: WidgetManifest.isContainer flows into the registry.",
"needs_flag": "carriers"
},
{
"at": "packages/types/src/base.ts ComponentMeta.isContainer ('Whether the component can have children'); packages/types/src/widget.ts WidgetManifest.isContainer; packages/types/src/zod/base.zod.ts ComponentMetaSchema.isContainer ('Can have children'); skills/objectui/guides/plugin-development.md ('Accepts child components'); content/docs/guide/react-pages.md",
"role": "PUBLISHED authoring declarations and docs of the flag's meaning",
"needs_flag": "They declare the flag to plugin authors. Under C they become declared-but-unenforced for containment unless they are re-described or retired, which is a contract decision."
},
{
"at": "apps/console/src/tests/public-contract.test.ts DECLARED_LAYOUT_CONTAINERS (getMeta(type)?.isContainer === true)",
"role": "test reader (objectui#6879 PUBLIC_BLOCKS ledger)",
"needs_flag": "yes, in its layout-containment meaning (same meaning as react-page)"
},
{
"at": "packages/app-shell/src/views/metadata-admin/previews/PageBlockCanvas.tsx local const isContainer, computed from childGroups(block).length (L710)",
"role": "local variable of the same name",
"needs_flag": "NOT a flag reader (confirmed: computed from the block's child groups)"
},
{
"at": "packages/sdui-parser/verify.ts fixtures; scripts/check-lucide-icon-record-names.mjs (header comment only); scripts/container-declaration-baseline.json note",
"role": "dev script / prose",
"needs_flag": "no"
}
],
"registrations": {
"instrument": "Temporary probes, never committed: the ratchet's own runtime predicate (SchemaRenderer with one authored text child; did it reach the DOM), extended over every KNOWN key, with the tier's manifest built the way getJsxManifest builds it. Population A: packages/components only, which is the ratchet's population: 295 keys, 132 bare. Population B: the 16 eager modules apps/console/dev/manifest-dump.tsx loads, plus @object-ui/layout: 498 keys, 279 distinct registrations, 0 lazy, 0 unknown-component. A third probe re-ran the blind cases inside their provider context.",
"full_registry_distinct": "74 registrations render authored children. 10 publish a children input the tier can see: page:card, page:footer, page:section, page:sidebar, ui:app, ui:home, ui:page, ui:record, ui:utility and ui:tooltip; tooltip is probe-blind (portal) and renders when open. 65 render children with NO child-list input: 15 of them only via isContainer, 50 via nothing.",
"underivable_flagged_15_bare": "article, aside, aspect-ratio, box, card, container, flex, footer, grid, header, main, nav, page-header, section, stack. If the flag stopped deciding containment and the tier read inputs, each of these would start drawing a FALSE not-a-container, making the diagnostic weaker. The ruling forbids that.",
"underivable_unflagged_50_bare": "the 44 baseline undeclared tags minus the 5 page kinds (39), plus the 11 excluded (button, alert, badge, 8 sidebar-*). All ten converged registrations are here.",
"probe_blind_extra": "measured in provider context: sidebar TRUE, sidebar-menu-button TRUE, tooltip(open) TRUE, sidebar-trigger FALSE (control, it reads no children). So sidebar and sidebar-menu-button are a further under-count beyond comment 5732585445: they draw a false not-a-container, appear in neither baseline list, and are underivable. The ratchet's comment calling them not part of the cost is false for two of its three.",
"flag_but_probe_negative": "page:accordion, page:tabs, responsive-grid, app-schema-renderer, layout:page:card. These declare isContainer but put no authored text child on the page under the bare probe. NOT MEASURED in context; in scope of the same derivation.",
"zone3_mechanism_falsified": "Slot inputs cannot be the signal: 0 of the 65 underivable declare any slot input, and 10 registrations that read no children (alert-dialog, collapsible, context-menu, dialog, drawer, dropdown-menu, empty, hover-card, popover, sheet; their sources contain no schema.children) declare slot inputs, so a slot-based derivation would silence a TRUE diagnostic. The body input cannot be the signal either: the only registration declaring it is record:alert (string|object translation map), which renders no child list. Only an input NAMED children is sound, and 10 registrations carry one.",
"baseline_reconciliation": "The ratchet and census are green at 62597c5 (2 files, 35 tests), so undeclared 44 + excluded 11 = 55 exactly equals today's runtime violators (render children, no flag). tooltip, sidebar and sidebar-menu-button are in neither list (probe-blind). The 5 page kinds sit in undeclared yet publish a children input (the page.tsx pageMeta name: 'children', type: 'array'), so they are the only baseline rows a derived reading can already clear."
},
"stop_condition": "FIRES on both arms: a reader outside the tier needs the flag (react-page; objectstack save gate), and 65 registrations (+2 context-only) have a child-list reading the tier cannot derive. Per the ruling, no fallback to the flag was invented; the route is returned for decision below."
},
"tests": "All heavy runs went through os-verify-lock (slot issue-9910-dev), each reading VERDICT command-exit 0. (1) pnpm exec vitest run packages/components/src/renderers/__tests__/zz-9910-step1-probe.test.tsx → 1 passed, 295 keys dumped. (2) pnpm exec vitest run apps/console/src/__tests__/zz-9910-full-probe.test.tsx → 1 passed, 498 keys. (3) pnpm exec vitest run packages/components/src/renderers/__tests__/zz-9910-blind-probe.test.tsx → 1 passed, the context readings above. (4) Control at BASE: pnpm exec vitest run packages/components/src/renderers/__tests__/container-declaration-ratchet.test.tsx packages/components/src/renderers/__tests__/container-declaration-census.test.tsx → Test Files 2 passed, Tests 35 passed. All three probe files were deleted and never committed: git status --porcelain shows 0 lines at HEAD 62597c5. No ablation: no repair was written.",
"gates": "Nothing is owed because there is no diff (pr null). Each of these reads NOT MEASURED, reason no diff: sdui-parser type-check, components type-check, sdui-parser tests, check:sdui-registration-pins, check:control-bytes, check:new-line-citations. dispatch-gates.mjs was not run: it refuses this repo, as the claim recorded.",
"deviations": "none. No exported symbol, published meta type field or file changed. The branch was pushed EMPTY as the write-route probe at 62597c5 (git push, no commits beyond main); the remote ref cannot be deleted from this container (objectui AGENTS.md, ref-delete 403).",
"files_changed": [],
"mcp_calls": "0",
"api_writes": "1 — POST /repos//issues/9910/comments (this report, via scripts/pm/post-stamped.mjs and the fleet relay). The git push of the empty branch is not a REST write.",
"open_questions": [
{
"question": "Q1. How does a registration's child-list reading reach the tier? Today 65 child-rendering registrations (+ sidebar and sidebar-menu-button in context) publish it nowhere the tier reads, and 15 of them publish it only through isContainer.",
"options": [
"A · declare-and-pin. Every renderer that reads schema.children declares { name: 'children', type: 'slot' }, the shape 10 registrations already carry (tooltip, page:card/footer/section/sidebar, the 5 page kinds). The tier's containment reads only that input. The existing runtime census is re-pointed from the flag to the input in both directions: the red-when-fixed pins flip to assert the new truth, and the excluded/undeclared rows are deleted as they clear. Four axes. ① Business: covers the 55 runtime violators plus tooltip, sidebar and sidebar-menu-button, which is the measured pull. ② Long-term: one published field (inputs) is read by the tier, the save gate and the docs table. It is still a declaration, held by an executed census rather than derived from code. ③ AI-error: the diagnostic stops lying on the only spelling that renders, with no flag fallback. A slot input emits no JSX attribute (read from codegen.ts emitInterface), so enforcement stays diagnostic-only. ④ Startup: no new field, no new gate (it re-points the existing ratchet). Cost: about 67 mechanical registration edits across components and layout, well outside the claimed file surface; the inputs of 9 public blocks grow in sdui.manifest.json (effect on registry-inputs-spec-parity and objectstack check:react-declaration-parity NOT MEASURED); and a lockstep port in objectstack (Q3).",
"B · build-time derived field. Manifest generation runs the runtime census and publishes a derived accepts-children field. ① Business: same coverage minus the probe's blind spots. ② Long-term: derived from executed code, which is closest to the ruling's 'cannot drift'. But it adds a field to the published Manifest/ManifestComponent type and a DOM render step to both manifest generators; objectstack's runs under plain Node. ③ AI-error: the instrument is measured blind to context and portal registrations (tooltip, sidebar, sidebar-menu-button), so it needs hand-kept context fixtures, which is a hand-kept list again. ④ Startup: a new mechanism plus a new published field. The heaviest option.",
"C · the flag keeps deciding containment and the debt is paid by declaration (the card's original A/B space). Declaring the flag on this population is refused by objectui#6804. Accepting the false diagnostic (the card's B) was held to be a transition at most. Listed only as the state that holds until Q1 is ruled."
],
"recommendation": "A. It is the only option that covers the measured population without a new published field or a DOM build step. It also dissolves the objectui#6804 conflation instead of papering over it: the children input means 'reads a child list' (button, badge, alert and tooltip declare it), and isContainer is left meaning only 'layout container' (Q2). Its residual risk, a hand-declared input drifting from the renderer, is held in both directions by the census that already exists. B is purer on axis ② but fails on ③ (blind spots come back as a hand list) and on ④."
},
{
"question": "Q2. What does isContainer become once it stops deciding containment? react-page.tsx needs a LAYOUT-containment signal that differs from the child-list reading exactly on button/badge/alert, and the flag is a PUBLISHED authoring field: ComponentMeta.isContainer, WidgetManifest.isContainer (the third-party widget path through WidgetRegistry), ComponentMetaSchema ('Can have children'), and the published skills plugin guide ('Accepts child components').",
"options": [
"A · keep the field and re-describe it everywhere as 'layout container: excluded from the react-page scope and the public-contract layout ledger', no longer 'can have children'. Drop it from, or re-describe it in, the SDUI manifest. ① Business: react-page keeps Button/Badge/Alert. ② Long-term: one meaning per field. ③ AI-error: a plugin author reading the guide is no longer told the flag admits children. ④ Startup: text-only on types/zod/docs. It touches skills/** (governed), so that part is Tier H.",
"B · retire the field; react-page switches to an existing signal such as category 'layout'. Whether that set equals today's flagged public set is NOT MEASURED. It removes a published field under the 'retire immediately' default, but it is only safe once that equality is measured.",
"C · keep the field and its current description. It becomes declared-but-unenforced for containment, against 声明即强制. Not recommended."
],
"recommendation": "A now. Take B only as a follow-up, and only if the category-layout equality is measured."
},
{
"question": "Q3. Cross-repo lockstep. objectstack's save-gate copy of the parser decides not-a-container from the manifest's isContainer, and its lockstep gate cannot see a predicate change (it compares codes and the parse.ts region only). Is the objectui change allowed to land before the objectstack port?",
"options": [
"A · port in the same objectui release and pin bump. One objectstack port card is filed, carrying a Blocked-by: line on this card. No window where saving and rendering disagree.",
"B · objectui first, port later. There is a window where the save gate warns (or stays silent) where the renderer does the opposite: the surface-dependent verdict objectstack#12719 exists to prevent."
],
"recommendation": "A. The objectstack#12719 invariant is exactly this case, and its mechanical gate is blind to it."
}
],
"out_of_scope_findings": [
"none filed. Every finding of this run sits inside this card's derivation (the ruling put tooltip, the page kinds and the whole containment population in scope), so they are in step1 rather than here. Dedupe words, should the seat split Q3 into an objectstack card: sdui-lockstep · not-a-container · isContainer save gate · validateJsxPages manifest containment"
]
}

---

Generated by Claude Code

[objectstack-fleet]

🗳️ Back to the decision box — ruling C's precondition was falsified at step 1 (one question left)

Release: session_01BA3nKVUwKQJf8DBxrSVtNC · cause: the ruled step-1 measurement fired the ruling's own stop condition on both arms (report 5808588768) · destination: needs-user-decision (director seat), assignee cleared in the same act
Ruling-ref: 5748934805 (letter C stands in DIRECTION — the containment diagnostic must follow what a registration really reads; only its execution precondition failed)
Governing text: objectui#6804 ruling 5468762497 「isContainer 的语义是布局包含,不是「这标签会渲染 children」」 · ruling 5748934805 「if a reader outside this tier needs the flag, or a registration's reading is underivable, the dev stops and reports — ⛔ no fallback to the flag is invented silently」 · docs/NORTH-STAR.md 优先级 4 「元数据既要 AI 能写——错的必须被响亮拒绝并给处方」 · objectstack#12719 (save gate and renderer must speak one dialect)

What the seat verified itself on origin/main before routing this (⛔ not relayed)

claim that decides the verdict	reading
a reader outside the tier needs the flag	objectui packages/components/src/renderers/layout/react-page.tsx still does if (!tag || cfg.isContainer) continue; — the flag decides which blocks become JSX identifiers on every kind:'react' page
a cross-repo reader needs it too	objectstack packages/sdui-parser/src/validate.ts:168 (c1641868a3) has the same !comp.isContainer predicate, fed by ctx.sduiManifest from packages/lint/src/authoring-rules.ts (validateJsxPages) — the save gate
the child-list reading is not published	a name: 'children' input exists at only 4 declaration sites under packages/components/src/renderers/** + packages/layout/src/** (containers.tsx ×2, page.tsx, tooltip.tsx) — consistent with the dev's 10 registrations out of 74 that render children

⇒ the report is accepted as a measurement: stop condition fired, no code, no PR. The dev's branch claude/issue-9910-containment-from-child-reading exists on origin as an EMPTY branch (== 62597c5) that this container cannot delete; ⛔ it is not a live claim.

维护者速读 · 一句话问题

作者在 Studio 里写出一个能正常渲染的侧边栏、按钮、卡片,平台却提示「这里不能放子节点」。裁决 C 说「告警要跟着组件实际读不读子节点走」;执行前的测量发现:74 个会渲染子节点的组件里,只有 10 个把「我读子节点」写进了平台读得到的清单,其余 65 个只存在于代码里。所以要你裁的是:「我读子节点」这件事,以什么形式写进组件清单?

选项 × 真实代价

字母	做什么	作者 / 宿主看到的后果	代价(实测标注)
A 声明 + 普查钉住	每个会渲染子节点的组件在清单里声明一个 children 插槽输入(10 个组件今天已是这个形状);告警只读这个输入;现有运行时普查改为双向核对「声明 = 实际渲染」	假告警在全部 74 个组件上消失;真不收子节点的组件照样报	约 67 处注册改动(components + layout);9 个公开块的清单输入变多(对两道一致性检查的影响 未测)
B 构建时推导	生成清单时实际渲染一遍,把「收子节点」作为新字段发布	覆盖面同 A,但 tooltip / sidebar 这类要上下文或弹层才渲染的组件,探针看不见(已实测看不见),要手工补清单	清单类型新增一个已发布字段 + 两个清单生成器各加一步 DOM 渲染(objectstack 那边跑在纯 Node)
C 维持现状	旗标继续决定告警;假告警照旧,已被钉成会变红的断言	作者继续在唯一能用的写法上看到假告警	零改动;缺陷留着

业务含义直译

• A = 每个组件在「说明书」里写清「我有一个放内容的槽位」,并且有一道每次都跑的检查保证说明书和实物一致 —— 像 Plasmic / Builder.io 把 children 当作一个声明出来的 slot 属性。
• B = 不写说明书,每次出厂前把实物拆开量一遍再印说明书;量不到的角落只好手抄。
• C = 说明书继续写错,但在错处贴了一张「已知错误」的标签。

四轴论证(业务立场)

• ① 项目长远合理性 —— 两年后该有的样子:组件清单是每个块唯一的机器可读契约,「有没有子节点槽位」是清单里的一条输入,告警、文档表、objectstack 保存闸门三处读同一条,运行时普查保证它不和代码分家。A 正是这个终态;B 多一个平行字段加一套只在构建期存在的推导;C 保留已经分家过一次的那个旗标。
• ② 实际业务拉动 —— 今天就撞上:十个已收敛组件 + tooltip + 5 种页面 + sidebar / sidebar-menu-button(后两个是这次新量出的漏数)在 children 上报假告警。A、B 都覆盖,C 不覆盖。
• ③ 防 AI 犯错 —— AI 作者照文档写 children,平台今天回它一句假拒绝,它就会退回错误写法。A 之后:写对不报,写错(给不收子节点的组件塞子节点)照样响亮报;声明与实物一旦分家,普查当场变红。B 的盲区靠手抄清单补,手抄正是这个类别三次复发的机制。
• ④ 创业阶段不扩散 —— A 不加新字段、不加新门禁(改指向现有普查);B 加一个已发布字段和一步构建渲染;C 零改动但缺陷永久在。

os-decision-facets

• ① 长远合理性:A 让「收不收子节点」只有一个声明处并被普查钉住,缩小特例;B 新增平行字段;C 保留已分家的旗标。
• ② 业务拉动:今天约 18 个组件在唯一能渲染的写法上报假告警,作者在 Studio 里直接看到。
• ③ 防 AI 犯错:A 写对不报、写错响亮报、声明漂移即红;B 的盲区回到手抄清单;C 继续对正确写法说谎。
• ④ 不扩散:A 零新字段零新门禁;B 一个永久已发布字段加构建期渲染;C 零改动零收益。

Prior rulings read: iscontainer,not-a-container,container-declaration,children-input,sdui-lockstep → 0 hits; none; thread: 1 ruling(s) (5748934805); repo: objectstack-ai/objectui

推荐

荐 A,回退 C(维持钉住的现状直到 A 落地);⛔ 不荐 B。

只看①选 A;②③④ 是否翻转:否 —— ② 两者覆盖相同,③ 与 ④ 都更偏 A。
置信缺口:① A 让 9 个公开块的清单输入变多,对 registry-inputs-spec-parity 与 objectstack check:react-declaration-parity 的影响未测;② 「slot 类型输入不生成 JSX 属性」是 dev 读 codegen.ts 得出的,未执行;③ 5 个声明了旗标却在探针下不渲染子节点的组件(page:accordion、page:tabs、responsive-grid、app-schema-renderer、layout:page:card)在上下文里未测。
⚠️ 本推荐与 dev 的推荐同字母,但理由按上面四轴独立写出,⛔ 不是沿用。

已由既有裁决定下、⛔ 不再请你裁的两件(执行参数)

1. isContainer 以后是什么意思 —— objectui#6804 裁决 5468762497 已定:它是布局包含。所以已发布的描述(ComponentMeta.isContainer「Whether the component can have children」、ComponentMetaSchema「Can have children」、发布技能指南「Accepts child components」)与裁决不一致,执行时照裁决改写;其中 skills/** 的一段属受管面,单独开 PR 等人批。
2. objectstack 保存闸门同步 —— objectstack#12719 的不变量是「保存闸门与渲染器说同一种方言」,而现有 lockstep 检查看不见谓词变化。⇒ 本席定为同一次发布、同一次 pin bump 一起落:裁 A 后立一张 objectstack 移植卡,带 Blocked-by: 回链本卡。这是任务间顺序,属 PM 裁量。

裁后执行段(你只裁方向)

• 裁 A ⇒ 本卡回 pm:queue,重派时文件面扩到全部约 67 处注册 + 告警谓词只读 children 输入 + 普查双向改指 + 上面两件执行参数;objectstack 移植卡同笔立。
• 裁 B ⇒ 同上,但以新增已发布字段与两个生成器的渲染步为交付;盲区组件的手抄清单作为已知代价写进 PR。
• 裁 C ⇒ 本卡转 pm:on-hold,Restart-touch: 指向 packages/sdui-parser/src/validate.ts,假告警维持钉住状态。

_{domain:ui seat #1 · session_01BA3nKVUwKQJf8DBxrSVtNC · review of report 5808588768 + decision routing · 2026-09-24T06:03Z}

---

Generated by Claude Code