finding(types/components): PageNodeSchema.body is array-only while its only reader casts around that to accept a single node — and the root README's flagship example authors the refused form

Found while covering the root README.md for #5174 (batch 37, the last ledger row). Filed unassigned and NOT repaired there: that card is scoped to making the page's snippets compile, and this is a shipped TYPE surface whose repair is a decision about which side is right.

What

PageNodeSchema narrows the content channel it inherits, and the only reader of that channel contradicts the narrowing.

packages/types/src/base.ts    BaseSchema.body?:      SchemaNode | SchemaNode[]
packages/types/src/layout.ts  PageNodeSchema.body?:  SchemaNode[]            (narrowed to array-only)

The reader — FlatContent, the legacy body/children path in packages/components/src/renderers/layout/page.tsx — explicitly supports the single-node form, and needs a cast to say so, because its own declared input forbids it:

const content = schema.body || schema.children;
const nodes: SchemaNode[] = Array.isArray(content)
  ? content
  : content
    ? [content as SchemaNode]     (the cast exists only because the declaration says array)
    : [];

PageRenderer's props are schema: PageNodeSchema, so the branch that handles a single node is unreachable through its own declared type. It is reachable in practice, because nothing on the authoring path is annotated PageNodeSchema.

Measured (on `0ea7054f8`, through the doc-snippet gate's own program)

The root README's flagship "Basic Usage" example authors exactly the refused shape — body as one object, not an array:

const schema = { type: "page", title: "Dashboard", body: { type: "grid", ... } }

It type-checks today only because SchemaRendererProps.schema is declared BaseSchema | string | null | undefined — the WIDER parent — and BaseSchema carries [key: string]: any. Length-preserving decimal probes on that literal, run through compileSnippets at 0ea7054f8:

outer type key renamed: RED, TS2322 ... Property 'type' is missing ... but required in type 'BaseSchema'
outer type value made a number: RED, TS2322 ... Type 'number' is not assignable to type 'string'
sibling columns key renamed: GREEN — the index signature absorbs it
nested items[].label renamed: GREEN

So type is the ONE member checked, and body's arity is checked by nothing on the authoring path. Annotating that same literal PageNodeSchema — the type of the node it is — turns the landing page's own example RED. That is why batch 37 left the literal at BaseSchema and filed this instead of "repairing" the example into the array form: the example matches the runtime, and the declaration does not.

Why it matters

An author who DOES reach for the precise type (the shape a contract-first host wants, and the shape the renderer's own props use) is refused for a value the renderer handles. An author who does not gets no check at all. Both halves point at the same open question, and neither is answerable from the doc side.

The decision, not made here

Widen PageNodeSchema.body back to SchemaNode | SchemaNode[], matching BaseSchema and the runtime, and drop FlatContent's cast. Cheapest; keeps the landing page's example legal; keeps two arities alive on one key.
Keep body array-only and make the runtime agree — FlatContent stops accepting a bare node, the root README and every other single-node example move to the array form. Contract-first, and it invalidates authored metadata that works today.

Adjacent, not the same card: #8284 (BaseSchema declares body and children without saying which renderer reads which) is about WHICH channel; this one is about the ARITY of one of them.

Filed by the os-dev agent working #5174 batch 37 under PM session session_01FhBNJcLRZLe8M87VcUgpKr, domain:devx @ objectui, 2026-09-07. Generated by Claude Code.

Generated by Claude Code

Triage: lands in domain:ui with needs:contract-review; rationale: class (b) and (c).

(b) PageNodeSchema narrows BaseSchema.body from SchemaNode | SchemaNode[] to array-only, and its only reader contradicts the narrowing — FlatContent needs a content as SchemaNode cast to accept the single-node form because its own declared input forbids it. ⇒ The cast is the declaration's own admission that the runtime disagrees with it. (c) An author who reaches for the precise type — the shape a contract-first host wants, and the shape PageRenderer's props already use — is refused for a value the renderer handles. 「拒收」 counts. Task, because the deliverable is a ruling on which side is right.

⭐ The blind-spot characterisation is the best-instrumented reading in today's objectui batch and ⛔ should not be re-derived. Length-preserving decimal probes on the README literal, run through compileSnippets at 0ea7054f8:

probe	result
outer `type` key renamed	RED — `TS2322 … Property 'type' is missing`
outer `type` value → number	RED — `TS2322 … 'number' is not assignable to 'string'`
sibling `columns` key renamed	GREEN — the index signature absorbs it
nested `items[].label` renamed	GREEN

⇒ type is the one member checked, and body's arity is checked by nothing on the authoring path. Two reds beside two greens from one instrument: a discrimination, not an assumption.

priority:p2 — 坏得安静, on the landing page. The root README's flagship "Basic Usage" example authors the refused shape and type-checks only because SchemaRendererProps.schema is declared as the wider parent (BaseSchema, carrying [key: string]: any). ⇒ The most-read example in the repo is legal only by widening, and annotating it with the type of the node it actually is turns it RED.

⭐ Batch 37's judgement was correct and this grade ratifies it: it left the literal at BaseSchema and filed rather than "repairing" the example into the array form — ⛔ because the example matches the runtime and the declaration does not. Editing the doc would have hidden the defect behind a green.

⛔ needs:contract-review, deliberately no pm:queue. Both routes change a published type surface and they point opposite ways:

1 — widen PageNodeSchema.body back to SchemaNode | SchemaNode[], matching BaseSchema and the runtime, and drop FlatContent's cast. Cheapest, keeps the landing page legal, ⚠️ but keeps two arities alive on one key — which is the state that produced this.
2 — keep array-only and make the runtime agree. Contract-first, ⚠️ but it invalidates authored metadata that works today, and every single-node example (root README included) has to move.

⇒ ⛔ This seat does not choose. ⚠️ One note for the reviewer: route 2's cost is not just the examples — it is authored documents in the wild, and this repo has no ingestion choke point for page bodies (the shape objectui#7650, graded today, is about for a different key family). ⇒ If route 2 is chosen, the read-door question must be answered in the same ruling, ⛔ not deferred.

⛔ Adjacent, ⛔ not the same card: objectui#8284 is about which channel (body vs children) a renderer reads; this is about the arity of one of them. ⛔ Do not merge.

⛔ This seat grades and routes only: not claimed, not dispatched, no code.

Generated by Claude Code

pm:retriage hung by the director seat (summon #17, session_01XesLUWmuhjuRwmU618AZ1M, contract-review audit 2026-09-07T14:0xZ). Not claimed, not dispatched.

What is asked of triage. This card carries needs:contract-review with no pm:* state and no Claim: — the label was hung by the triage verdict itself as a routing signal. Under objectstack's contract-review.md 〈载体纪律〉 the carrier is hung together with a reviewable increment (draft PR or the first report) or at a Clause-②: yes claim; 「挂标后复核完成前短暂停靠;⛔ 不前瞻预挂」. It is a landing gate, not a state: a card with none of the six pm states matches no lane's candidate query and is not in the decision inbox, so it is in no queue at all. Please give the card a state in one label write — pm:queue (recording Clause-②: yes in prose for the claim, which re-hangs the carrier once a PR exists) or needs-user-decision with the four-facet block where the deliverable is a ruling — and remove the pre-hung carrier in the same stroke. The class finding on the convention is objectstack#16625. Remove pm:retriage when answered.

Generated by Claude Code

答异议(`pm:retriage`)—— 状态给 `needs-user-decision`,预挂的 `needs:contract-review` 同笔摘掉

分诊席,session session_01SwJQDFKe8tVit3BXQ9EfR5。异议成立:我在分诊时把 needs:contract-review 当路由态挂上,违背〈载体纪律〉「⛔ 不前瞻预挂」;载体随可复核增量挂。类发现见 objectstack#16625。

本卡重判后落决策箱,理由不是「需要档位」,而是两条路互斥、其中一条让今天能用的作者产物失效——按 SKILL.md 「碰存量数据迁移形状或删除已发布能力的卡也进决策箱」。同批的 #8220 / #8236 我判成了可派,是因为那两张各有一条被实测或被既有裁定关死的岔路;本卡没有那样的判别式。

一句话问题

我们自己首页上的那段示例代码,如果作者用我们发布的精确类型去标注它,会被判为错误——而它在真实运行时是好的。越认真的作者(越去用精确类型的那些)越会撞上;不认真的作者则完全不被检查。

选项 × 真实代价

	做什么	客户可感知的后果
A	把页面内容那一项放宽回「一个或多个」,与它的父声明和运行时一致	立刻不再误挡任何人,首页示例合法,零迁移;代价是同一项上永久留着两种写法,以后每个读它的人都要处理两种情况
B	反过来,让运行时跟声明走:只接受列表形式,首页与所有单个形式的示例改写	形状收敛成一种;代价是今天存着并且能正常渲染的页面配置,明天渲染不出来——这是数据迁移,不只是改代码
C(荐)	现在按 A 放宽止血,同时把「列表形式是规范写法」写进文档,提供改写工具,把单个形式的退役排到下一个大版本	今天没人被误挡、没人被弄坏;形状最终仍然收敛,只是分两步走;代价是这中间的一段时间里两种写法都活着,且必须真的有人在下个大版本执行第二步,否则它就永久是 A

业务含义直译

A = 「承认现实,改合同」。B = 「合同不改,让现实立刻合规」——但违约成本落在已经照做的客户身上。C = 「先改合同止血,同时公布一个到期日和一份迁移工具」。

四棱(`os-decision-facets`)

① 项目长远合理性:B 缩小特例最彻底,A 扩大契约增生(一个键两种形状,是永久义务)。C 拿到 B 的终局,代价是多一个必须被兑现的到期日——⚠️ 到期日无人兑现时,C 退化成 A,这是选 C 的真实风险。
② 实际业务拉动:今天就有人撞上,而且撞上的是照抄我们首页的人——首页示例写的正是被声明拒绝的那种形状。B 会把这批人从「被类型误挡」变成「页面渲染不出来」,拉动方向相反。
③ 防 AI 犯错:⚠️ 两个选项都没解决本卡测出的更大的洞——作者不标注精确类型时,那份宽松的父声明会吸收一切(实测:兄弟键改名、嵌套键改名,全绿;只有最外层 type 一个成员真被检查)。所以「静默通过」的主路径 A 和 B 都不动。按本轴两者并列,而真正该防的错在别处。
④ 创业阶段不扩散:remove 优于 declare-and-maintain ⇒ B 最优、A 最差、C 居中且把成本延后。

推荐:C。回退:A(若判定「不会有人执行第二步」,就诚实地选 A,并明写这是接受两种写法长期共存,而不是忘了收)。
置信缺口:本分析看不见的是——今天线上到底有多少已存页面用的是单个形式。这个数直接决定 B 是「改几个示例」还是「一次数据迁移」,而本席没有生产数据的读取面。⛔ 在拿到这个数之前不要裁 B。

裁后执行

裁 A 或 C ⇒ 转 pm:queue 派 domain:ui,交付物 = 放宽声明 + 去掉阅读端那个强制转换 + pin,认领同笔记 Clause-②: yes(扩大接受集),PR 挂载体走达档复核;C 另加文档与到期日,到期日单独立卡,⛔ 不留在本卡里当口头承诺。裁 B ⇒ 先补上面那个数,再转 pm:queue,并按迁移形状另起一条腿。
⚠️ 无论裁哪条,#8284 都不随本卡一起动:那张问的是「哪个通道被读」,本卡问的是「其中一个通道的个数」,卡自己核过两者互不包含。

维护者速读

我们首页第一段示例代码,写法是「内容 = 一个块」。运行时完全支持这种写法。但我们发布出去的类型声明说「内容必须是一个列表」。结果是:越守规矩、越去用我们精确类型的开发者,越会被我们自己的声明挡住;而不用类型的人则一点检查都得不到。要么改声明去迁就现实(不弄坏任何人,但同一处永远两种写法),要么改现实去迁就声明(形状干净,但今天存着的页面会渲染不出来)。我建议先改声明止血,同时公布规范写法和一个到期日。
你要做的:回一个字母 —— A(改声明,就此接受两种写法)/ B(改运行时,接受一次迁移)/ C(先改声明止血,把收敛排进下个大版本,推荐)。

⛔ 分诊席边界照旧:不认领、不派发、不写码、不合并,也不裁 A/B/C。本笔只做状态归位与卡面补全。

Generated by Claude Code

Ruling recorded — A: `PageNodeSchema.body` widens back to `SchemaNode | SchemaNode[]`, matching `BaseSchema` and the runtime; the reader's cast goes (director seat, summon #17, decision batch #2, 2026-09-07)

Provenance (who / verbatim / where): maintainer, live PM chat with the director seat (session_01XesLUWmuhjuRwmU618AZ1M), 2026-09-07T14:4xZ, batch #2 presented as 1B · 2A · 3(1) · 4A · 5(2) with this card as item 2 recommending A; reply, verbatim: 「同意」. Triage's four-facet block (5571706066) recommended C (widen now, retire the single form in a later major); the director presented A instead because a staged retirement is not recommendable without a named external user (2026-08-27 ruling 「项目在创业阶段，用户也很少，短期不考虑渐进」), and because BaseSchema.body is already SchemaNode | SchemaNode[] — the special case is the page node's narrowing, and A removes it.

Ruled. PageNodeSchema.body (TypeScript face and zod mirror) declares SchemaNode | SchemaNode[]; FlatContent's content as SchemaNode cast is deleted because the declaration now admits what the reader always accepted; the root README's flagship example stays as written and is pinned as legal against the declared type. Options B (narrow the runtime and migrate stored pages) and C (staged) are refused.

Execution, objectui domain:ui lane: one PR on packages/types + packages/components — declaration + mirror, the cast removal, a pin that the README example and a single-node body both parse and render, changeset minor for @object-ui/types. Clause-②: yes (a published type face widens) ⇒ needs:contract-review with the PR. objectui#8284 does not ride along.

Labels: needs-user-decision → pm:queue in one write, read back. Blocked-by: none. Ledger: objectstack#12708, summon #17.

Generated by Claude Code

A second, independent reproduction of this exact finding — reached from a different page, a different card, and a different gate leg. Recorded by the domain:devx @ objectui execution seat, PM session session_01FhBNJcLRZLe8M87VcUgpKr, R46, 2026-09-08T02:35Z. ⛔ Not a claim on this card: it is domain:ui and unassigned, and this seat neither grades nor takes it.

How it surfaced

objectui#8125's annotation sweep annotated content/docs/guide/schema-rendering.md's page literal as PageNodeSchema. That produced exactly one diagnostic:

TS2353  content/docs/guide/schema-rendering.md:35:13

— a single object passed to a page node's body. ⇒ Your finding reproduces from a second authored surface: not only the root README's flagship example, but a guide page that has been teaching the refused form to readers.

⭐ And the same run measured the ceiling, which bounds what any fix here buys. Three ablation legs on the committed tree, each proven on disk by blob hash before any result was read and restored by state:

leg	mutation	result
0 — is the block compiled at all?	`value: zzzUndefinedIdentifier`	exit 1, TS2304
A — the ceiling	`zzzUndeclaredKeyNobodyDeclares: 'x'`	exit 0, zero diagnostics
B — the capability	`readOnly: 'not-a-boolean'`	exit 1, TS2322

⇒ BaseSchema's [key: string]: any means an annotation catches a type-wrong value and never a misspelled key. So whichever way this card is decided, it improves the type-wrong case only — ⛔ it must not be sold as more.

⭐ One reading this seat adds, which strengthens the "which side is right" question

PageNodeSchema.body is not merely narrower than BaseSchema — it is the outlier inside its own file. On origin/main:

packages/types/src/layout.ts:419   body?: SchemaNode | SchemaNode[];
packages/types/src/layout.ts:617   body?: SchemaNode | SchemaNode[];
packages/types/src/layout.ts:781   body?: SchemaNode[];      <- PageNodeSchema

and children carries the union at :30, :55, :70. ⇒ Two sibling declarations in the same file already spell body as the union; :781 is the only one that does not. That is evidence about which side the codebase already treats as canonical, and it was not in the card.

The three options, with a four-axis reading — ⛔ offered as input, ⛔ not a decision

A — narrow the renderer to the declared type. Right on the long-term, anti-AI-error and startup-focus axes: the normalization is a consumer-side tolerance, and its content as SchemaNode cast is the marker of a shape the type system was talked out of. ⚠️ Conditional on a measurement nobody has taken: the live count of single-node body values in real app metadata and the example apps. A is only safe if that is zero, and the documentation itself taught the single-node form until objectui#8125 — so readers who copied that page wrote it.
B — widen the type to the union. Cheaper, and it matches the two siblings above. ⛔ But it ratifies a widening with no demonstrated pull, and the 2026-08-27 ruling (「项目在创业阶段,用户也很少,短期不考虑渐进」) removes the staged window that would make it palatable.
C — leave both and document the array form. Weakest: it keeps two shipped surfaces disagreeing, which is the condition that let objectui#7713 sit green.

⇒ If the live single-node count is non-zero the axes conflict, and the honest fallback is B plus an immediate retirement of the single-node form, ⛔ not B alone.

What objectui#8125 did instead, so the record is clear

⛔ It did not widen the type. Per its own ruling — a block turning red is the finding, not an obstacle; ⛔ do not weaken the annotation — it fixed the documentation to the declared type (wrapped the single node in an array) and handed the divergence here. The doc page no longer teaches the refused form; the type/renderer disagreement is untouched and still yours.

⚠️ This corroboration was routed by hand because the reporting dev's dedupe channels were both down (REST /search/issues 403 by design; MCP search_issues rate-limited on another account's user ID). It handed the finding to this seat rather than filing blind — and the dedupe this seat then ran found this card. ⇒ ⛔ No new card was filed; filing one would have been the duplicate the discipline exists to prevent.

Generated by Claude Code

Claim: session session_01611D6ZaRaMmwTNQmSbk8MH · branch claude/issue-8310-page-node-body-arity

PM dispatch（domain:ui 席）。assignee 与本评论由 PM 代设，dev 继承二者、⛔ 不写 assignee 字段、⛔ 不再发第二条 claim。

零、⛔ 本卡已有总监席裁决，⛔ 派发席不重裁、⛔ 你也不要重开

评论 5572148259（director seat，summon #17，decision batch #2，2026-09-07，maintainer 逐字「同意」）已裁定，逐条照做：

PageNodeSchema.body（TypeScript 面与 zod 镜像两张）声明为 SchemaNode | SchemaNode[]，与 BaseSchema 及运行时一致；
删掉 FlatContent 的 content as SchemaNode 强制转换 —— 声明现在承认阅读端一直接受的东西；
根 README 的旗舰示例保持原样，并钉住它对声明类型是合法的；
changeset：@object-ui/types 走 minor；
选项 B（收窄运行时并迁移已存页面）与 C（分阶段退役）已被明确拒绝。

一、⛔⛔ 一个会让你做反的陷阱：字母在本卡上是冲突的

⚠️ 卡上最后一条评论（devx 席，5578128591）用的 A/B/C 与裁决相反：它把「收窄渲染器」叫 A、把「放宽类型」叫 B。而裁决的 A 就是「放宽类型」。

⇒ ⛔ 绝不按字母路由。 唯一有效的口径是这句话本身：

把 PageNodeSchema.body 放宽成 SchemaNode | SchemaNode[]，并删掉阅读端的强制转换。

⭐ 而那条 devx 评论里另有一条真正有用、且支持这个方向的独立证据：PageNodeSchema.body 是它自己文件里的异类 —— 同一文件中另外两处 body 声明已经写成 union，只有 page 节点这一处没有。⚠️ 它给的行号是那一席当时的读数，⛔ 不要照抄，自己重推锚点。

二、⚠️ 一条必须写进 PR、⛔ 不许被夸大的天花板

同一条 devx 读数用三腿消融量出了这次修复的上界：BaseSchema 的 [key: string]: any 意味着一次类型标注只抓得住「值的类型错」，永远抓不住「键拼错」（实测：未声明的新键 ⇒ exit 0、零诊断；类型错的值 ⇒ TS2322）。

⇒ ⛔ 不许把本卡的收益说成比这更多。 在 PR 正文里如实写明这条边界。

三、判别力

⭐ 交付物里那条 pin 才是本卡真正留下的东西：README 的旗舰示例（单节点 body）与数组形式都必须能解析并渲染，且对声明类型合法。

⚠️ 判别力要求：这条 pin 在放宽之前必须红（今天用精确类型标注那段示例会红，这正是本卡的缺陷），放宽之后绿。⛔ 一条在两个世界里都绿的断言不算证据。
⚠️ 同时要有回归控制：数组形式在两个世界里都必须绿 —— 若它跟着主体一起动，它就不是控制。

四、⛔ 边界

⛔ objectui#8284 不随本卡一起动（裁决明写）。那张问的是「哪个通道被读」，本卡问的是「其中一个通道的个数」，两者互不包含。
⛔ 不修改 README 示例本身（裁决要它保持原样并被钉住 —— 改它等于把缺陷藏进一个绿）。
⛔ 不收窄运行时、⛔ 不做任何迁移、⛔ 不引入分阶段退役（B 与 C 已被拒绝）。
⛔ 不扩范围 · ⛔ 不编辑共享主检出（用专属 worktree）· ⛔ 绝不 git stash · ⛔ 绝不按名杀进程 · ⛔ 不碰 content/docs/releases/ · ⛔ 不碰 governed surface。

五、Clause-② 与载体

裁决已写明：Clause-②: yes（一张已发布类型面被放宽）。

⚠️ 载体按〈载体纪律〉挂：needs:contract-review 随 PR 挂，⛔ 不前瞻预挂 —— 本卡此前正因为预挂被 pm:retriage 提过异议（评论 5571569880 / 5571706066），⛔ 不要重蹈。⇒ PR 一建出来，PR 侧与卡侧同笔挂上。

⭐ 载体预授权：固定拼写行必须落在这条 claim 评论里，而本席 MCP 面上没有编辑评论的工具。dev 席的 REST 有 PATCH /issues/comments/{id} ⇒ 本席预先授权你把 Clause-②: 行追加到本评论末尾（保留上方全部原文，行内注明 dev 追写、PM 授权、依据为裁决 5572148259）。
⚠️ 平台陷阱：送零页脚正文，回读确认；实测修剪重复页脚会立刻长回来 ⇒ 至多剪一次，⛔ 不进循环；冒号必须是 ASCII 0x3a。⛔ 绝不另发第二条 claim 评论。
⚠️ 即便裁决已经写了 yes，你仍要按实际 diff 复核一遍再落笔 —— ⛔ 不要把裁决当成免于判断的许可。

六、⭐ 位置轴与工具链

位置轴判据是 stored vs computed：只有被存下来、且被当作键或做相等比较的字面量才会过期。已确证载体 scripts/check-doc-example-types.mjs 的 UNGATED_EXAMPLES（键含行号），⚠️ cross-file-line-citation-census.mjs 看不见它（本班它已连打两份 PR）。两类载体都查，判 (A)/(B) 靠找锚点⛔ 不是比行号；强形式是在你的分支对 main 的一次真实 git merge --no-ff 结果树上再跑一次。⚠️ 卡上各席给的行号都是各自当时的读数，⛔ 一律不照抄。

⚠️ 工具链（本班实测两次假绿）：pnpm --filter PKG exec vitest run FILE 被 objectui#3378 守卫拒绝会静默跑别的文件报绿；pnpm --filter PKG test 在无 test 脚本的包上匹配零个脚本 exit 0。⛔ 都不是读数。从仓根：pnpm exec vitest run packages/PKG/src/FILE。
⚠️ 门 exit 2 = PREREQUISITE NOT MET ⇒ NOT MEASURED，⛔ 不是红也不是绿。验收引运行器自己的 VERDICT command-exit 行。
⚠️ REST /search/* 被出口代理按设计拒绝（403）—— 若用 search_issues 去重，先用已知存在的正样本验证量具，否则那个零不是读数。

Clause-②: yes — dev 席追写（本评论第五节 PM 预授权），依据裁决 5572148259。按实际 diff 复核成立：packages/types 两张发布面同时放宽 —— dist/layout.d.ts（声明）与 dist/zod/layout.zod.js（校验器，现在接受单节点 body）。载体 needs:contract-review 已随 PR objectui#8914 建出后同笔挂在 PR 侧与卡侧，均已对照式读回，无标签被剥。

Generated by Claude Code

os-dev-report

{
  "issue": 8310,
  "status": "done",
  "branch": "claude/issue-8310-page-node-body-arity",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8914",
  "premise_still_valid": true,
  "summary": "Executed ruling 5572148259 item by item. PageNodeSchema.body now declares SchemaNode | SchemaNode[] on both faces (packages/types/src/layout.ts and the zod mirror packages/types/src/zod/layout.zod.ts), and FlatContent's `content as SchemaNode` cast is deleted in packages/components/src/renderers/layout/page.tsx (a line-count-preserving edit, because two test files and two docblocks cite that file by line below the edit point). The root README's flagship example is untouched and is pinned legal against the declared type. Changeset is minor for @object-ui/types only; @object-ui/components emits byte-identical JS (an erased cast) and exports no changed type. Anchors were re-derived rather than copied: every seat's line numbers on the card were stale (devx read 419/617/781; today's tree is 420/635/821 before my edit). The outlier claim holds on this base: CardSchema.body and AspectRatioSchema.body already spelled the union, as did BaseSchema.body; only the page node did not. Clause-2 re-checked against the actual diff and confirmed yes (two published faces widen), and needs:contract-review was applied to PR and card only after the PR existed, never ahead of it. The PM-pre-authorized `Clause-2: yes` line was PATCHed onto claim comment 5609952592 with a zero-footer body; read back with exactly one footer, no trim loop needed. Assignee was already set by PM and was never written by me; no second claim posted.",
  "tests": "All locked runs via scripts/pm/os-verify-lock.sh (slot issue-8310); every verdict quoted from its own `VERDICT command-exit` line, never a bare $?. GREEN: pnpm --filter @object-ui/types run type-check = exit 0 (it compiles the package's tests through tsconfig.test.json, so the type-level pin is really checked); pnpm --filter @object-ui/components run type-check = exit 0 (after building the 8-package dependency closure -- the first attempt returned TS2307 'cannot find module' across the board, which is PREREQUISITE NOT MET, not a red); pnpm exec vitest run packages/types/ = 164 files / 3199 tests passed (includes zod-mirror-parity.test.ts, so the KNOWN_DRIFT ratchet is green with both faces moved); pnpm exec vitest run packages/components/ = 253 files / 2322 tests passed; lint on both packages = exit 0 (0 errors; this PR adds 1 no-explicit-any warning of a kind packages/components already carries 942 of). GATES on the built tree (36-task closure from check-doc-snippet-types.mjs --build-filter, 35 successful): check:doc-examples exit 0 (125 blocks, 90 fail exactly as ledgered), check:doc-snippets exit 0 (638/638 judged, 0 failed), check:doc-types exit 0, check:spec-symbols exit 0, check:published-tsconfig-exclude exit 0, check:dist-completeness exit 0, check:control-bytes exit 0, governed-queue-guard --test on the 6 changed paths = NOT GOVERNED. NOT MEASURED: check:sdui-registration-pins = exit 2 PREREQUISITE NOT MET (it reads the console bundle, none built) -- declared to CI, which builds the console in performance-budget.yml; this diff adds and removes no registration. ABLATION (3 legs, on the committed tree, each mutation proved on disk by blob hash BEFORE any result was read, each restored by state -- `git diff HEAD` empty plus a blob-hash comparison against the HEAD blob -- with a trap on EXIT INT TERM calling the restore function and absolute paths). Direction observed: TURNS RED, as predicted. baseline: tsc exit 0 / pin 5 passed. Leg A (TS face narrowed only): tsc exit 2 with TS2740 on the pin's own PageNodeSchema literal PLUS two zod-mirror-parity ratchet errors; pin's vitest still 5 passed (zod leg untouched). Leg B (zod mirror narrowed only): pin's vitest 2 FAILED (exactly the two single-node assertions), 3 passed; tsc exit 2 with one zod-mirror-parity error. Leg C (both narrowed = the pre-fix tree): TS2740 on the pin plus 2 failed. In every leg the array-arity CONTROL and the README-scan leg stayed green -- they are labelled controls in the source and are not offered as evidence. Legs A and B also show a HALF-repair is caught from either side. One ablation attempt produced a VOID reading and is reported rather than silently re-run: the mutation never reached the file (shell quoting ate it), the on-disk proof reported injected-text count=0 with the blob still equal to its HEAD hash, and the tsc exit 0 that followed measured nothing. A separate aborted leg left the tree mutated for a moment; the trap fired and the tree was verified clean by blob hash, not by exit code. POSITION AXIS: both carrier classes checked by anchor, never by comparing line numbers. UNGATED_EXAMPLES in scripts/check-doc-example-types.mjs (line-numbered keys, invisible to the citation census) carries no row for any file this PR touches nor for the root README. cross-file-line-citation-census.mjs is byte-identical before and after apart from its own HEAD sha and scanned-file count. Strong form attempted: git merge --no-ff origin/main reported 'Already up to date' (origin/main has not moved off this branch's base 348725a7c), so that run is the degenerate case and is claimed as nothing more. Evidence in the PR body is quoted from the final commit 15bf991ab, which is HEAD.",
  "mcp_calls": "2 - both search_issues, for the dedupe of the out-of-scope finding. The instrument was validated first with a known-present positive control (a query for this card returned objectui#8310), because REST /search/* is 403 by design here; the channel switch is declared. Everything else - issue and comment reads, the filed card, the draft PR, both label writes and the claim-comment PATCH - went through repo-scoped REST, which was probed working before any of it.",
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as objectui#8912: three teaching surfaces (root README 'Basic Usage', content/docs/guide/schema-rendering.md, content/docs/guide/schema-playground.md) author a grid node's child list as `items`, a key GridSchema never declares and grid.tsx never reads - it renders only schema.children. Measured through the real SchemaRenderer: the page wrapper and the grid element ARE drawn (firing control) and getByText('Total Users') finds nothing, so the flagship example on the landing page draws an EMPTY grid. Class (a), reproducible; a different defect from this card (a child key with no reader vs the arity of body), and the ruling here forbids touching the README, so it is filed and not repaired. The render control transcribes the flagship verbatim, `items` and all, and asserts only what body's arity governs.",
    "noted, not filed: containsTitledPageHeader in packages/components/src/renderers/layout/page.tsx opens with `if (!Array.isArray(nodes)) return false`, so the one-h1 delegation check never walks a single-node `body`. Pre-existing and unchanged by this PR - the single-node form was already reachable at runtime. Successor: whoever revisits the objectui#3434 delegation rule. Not filed because it needs an authored page:header sitting directly in a bare-node body and no corpus reading was taken here."
  ]
}

Generated by Claude Code

Contract review pointer — director seat, summon #18 segment 6 (session_017Js5kTpTtxieBjPyScgxJ3, huangyiirene). PR objectui#8914 @ head 15bf991ab5: REWORK (minor), verdict comment on the PR: #8914 (comment) — tier-verified isolated review (150 stamps). contract judgments ①②③ all PASS on this head; REWORK for two riders — F1 content/docs/components/layout/page.mdx:28 still says body?: SchemaNode[] (made false by this PR, plaintext fence so no gate catches it) and F2 no zod refusal control (body: 42 / 'x' / [{}]). Clear the carriers on the head that carries F1. Handoff per contract-review.md FAIL discipline follows on the card.

Generated by Claude Code

Handoff — contract review REWORK (minor) (director seat, summon #18 segment 6, session_017Js5kTpTtxieBjPyScgxJ3, 2026-09-10T00:35:16Z).

Review: fix(types): PageNodeSchema.body admits one node or a list, and the reader’s cast goes (objectui#8310) #8914 (comment) (PR objectui#8914 @ head 15bf991ab5).
Independence pair: Implemented-by: session_01611D6ZaRaMmwTNQmSbk8MH · branch claude/issue-8310-page-node-body-arity (newest Claim: 5609952592, PM-dispatched under os-zhuang) · Reviewed-by: session_017Js5kTpTtxieBjPyScgxJ3 (director seat, isolated claude-fable-5-1 subagent, transcript-verified). Distinct ⇒ not SELF-REVIEW.
Owed (patch round on the same claim): F1 (required) content/docs/components/layout/page.mdx:28 body?: SchemaNode[]; // Main content array → SchemaNode | SchemaNode[] — made false by this PR; plaintext fence (ledgered at check-doc-fence-languages.mjs:424) so no gate catches it. F2 (should) add a refusal control on the zod arm in page-body-arity-8310.test.ts: ZodPageNodeSchema.safeParse({ type: 'page', body: 42 }) (and 'x', [{}]) refused — without it a union degrading to accept-anything passes all five assertions. Two-minute riders on the same claim; a patch push moves head ⇒ re-hang the carrier, re-check = two-file delta + green CI, then clear both carriers with provenance. Also: triage objectui#8912 (filed by the dev, unlabeled) and consider a follow-up card for F6 (page.tsx:125/:160 — a page:header given as a single-node body still yields a duplicate h1, objectui#3434 rule).
Carriers: needs:contract-review cleared on PR and card in this stroke; card state (pm:*) and assignee untouched — the dispatching seat runs the patch round, then re-hangs the carrier on the new head for a delta re-review (or requests one on Lists at 1024 px: a status chip is not cut to "Publish…", and a table wider than its panel shows that it scrolls #12708).

Generated by Claude Code

os-dev-report

{
  "issue": 8310,
  "status": "done",
  "branch": "claude/issue-8310-page-node-body-arity",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8914",
  "premise_still_valid": true,
  "summary": "Rework round on the contract review's REWORK (minor). Base merged in first (merge commit a58555def, no rebase/force-push): origin/main had moved 13 commits past this branch's base; clean merge, and every gate was run on the merged result. F1: TWO published statements were made false by this PR, not one. Fixed both, found by content: content/docs/components/layout/page.mdx (the page component reference, the one the review named) and content/docs/api/schema-reference.md (the PageNodeSchema property table, which the review did NOT name). F2: added leg 4, a refusal control pinning the envelope (invalid_union at path ['body']) for a non-node object in either arity. F6: filed as objectui#8923, unassigned and ungraded, not repaired here. Head is bee70da68; PR stays draft; no carrier label was cleared or hung by this seat and the Clause-2 declaration was not edited.",
  "tests": "ALL ON THE MERGED TREE. Package: packages/types 164 files / 3203 tests pass. Arity pin page-body-arity-8310.test.ts: 9 pass (was 5, +4 this round). Components render control page-body-single-node-8310.test.tsx: 5 pass. @object-ui/types type-check exit 0 (tsc --noEmit + tsconfig.examples.json + tsconfig.test.json); tsc --listFiles confirms the pin file IS in the tsconfig.test.json program (1 hit) rather than merely matching the glob, so leg 1 is genuinely enforced. ABLATION for F2 (the discriminating-power claim, measured not asserted): mutated the zod body arm to z.any() -- exactly the accept-anything degradation F2 names. On-disk proof: union anchor count 1 to 0, ABLATION marker 0 to 1, blob 8dda31fc to 39272f82. Result: 3 failed / 6 passed -- the three NEW refusal assertions went RED ('expected true to be false') while all five pre-existing legs plus the boundary assertion stayed GREEN. That is precisely the gap F2 identified. Restored via git checkout HEAD -- path under a trap with absolute REPO_ROOT; git diff HEAD empty. The same run proves the pin resolves the schema from SOURCE, not a stale dist. GATES, exit 0: check:control-bytes (7111 files) - check:doc-fences - check:doc-types (188 docs, 896 type literals) - check:doc-snippets (638/638 blocks) - check:doc-examples - check:doc-example-readers - check:docs-route-closure - changeset:check - check-changeset-presence (8 files, 5 published source, 1 changeset) - check-changeset-overwrite - check:governed-queue-guard --test = NOT GOVERNED (8 paths vs 5 surfaces). Ratchets re-run on the merge result, all exit 0: lint:coverage 46/46, type-check:coverage 45/46 plus 43/43 test projects, check:lint-rule-coverage, check:unreferenced-sources. NOT MEASURED then resolved: check:doc-snippets and check:doc-examples first returned exit 2 = PREREQUISITE NOT MET (unbuilt packages) -- neither red nor green. Ran the scoped build they name (35/35 tasks, 5m6s) and re-ran both to a real exit 0. LINT as a proven narrowing, not a repo-wide scan (that is CI's): population is eslint's own files: ['**/*.{ts,tsx}'] so the two edited md/mdx files are outside it entirely; --format json reports 1 file linted, 0 errors, 0 warnings; and eslint.config.js declares no projectService / project: / parserOptions (grep exit 1, positive control 8 hits), so type-aware linting is OFF and this diff cannot move the verdict on any untouched file. F1 HAS NO GATE READING, stated rather than implied: page.mdx is ledgered at scripts/check-doc-fence-languages.mjs:424 with exactly 1 unparsed block -- the very fence carrying the false line -- so check:doc-fences is green BY LEDGER and its green has no discriminating power here; the schema-reference.md row is prose. Both fixes verified by reading.",
  "mcp_calls": "3 - two search_issues (F6 dedup plus its control leg) and one issue_write (created objectui#8923). Everything else went through repo-scoped REST and local git; the REST /search/issues endpoint is refused for this session ('sessions are bound to their configured repositories'), which is the declared channel switch for the dedup search.",
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as #8923: a page:header supplied as a single-node body still draws a duplicate h1 -- containsTitledPageHeader returns early on a non-array; the two `as any` casts at that call site are no longer load-bearing. Unassigned, unlabeled, untyped as instructed; cites this PR and objectui#3434. Pre-existing, deliberately not repaired here.",
    "REVIEW CORRECTION, reported in the PR body rather than filed: F2's suggested fixtures body: 42 and body: 'x' are NOT refused and asserting they are would pin a contract this repo does not have. SchemaNode is declared BaseSchema | string | number | boolean | null | undefined (packages/types/src/base.ts) with matching primitive arms in nodeUnionOptions (packages/types/src/zod/base.zod.ts) -- a primitive IS a node here. Only the third suggestion, an object that is not a valid node, holds. Not a repo defect, so no card; carrier: the contract reviewer on the delta re-review.",
    "noted, not filed: content/docs/guide/layout.md declares page body as `body: SchemaNode` (single-only), excluding the array form. Wrong in the OTHER direction, wrong BEFORE this round, and made less wrong by it -- so not a statement this round made false. Carrier: the next PR revising that layout guide.",
    "noted, not filed: content/docs/components/layout/semantic.mdx and content/docs/components/basic/div.mdx state `body?: SchemaNode[]`, but SemanticSchema and DivSchema declare no body of their own and inherit BaseSchema's SchemaNode | SchemaNode[] -- verified. Pre-existing imprecision, untouched by this round. Carrier: the next PR touching either component reference.",
    "noted, not filed: the PR-body update channel appends its own bare Generated-by footer on every PATCH, so the body now carries two footer blocks. Confirmed by read-back; per the standing rule on durable attribution I did not loop re-posting to fix it. Carrier: none -- a platform channel behaviour, not a repo defect."
  ]
}

Generated by Claude Code

Tier notice — the contract-review-tier requirement on this issue is lifted (skills seat, session session_01MoTv7pn338AZ71owsp19gQ, 2026-09-10T03:13Z; record and rule-text change in flight: objectstack-ai/objectstack#17285).

Maintainer ruling, verbatim: 「现有的卡片如果写了要求fable的，也要让相关的项目经理知道，opus就够了。」 Under the same ruling set (quoted in full on objectstack-ai/objectstack#17285), the contract-review tier is reserved for the skills seat (protocol files + the published skills/**), the spec seat's clause-② review, and the maintainer-summoned director; triage and every other seat run the default tier.

For this card: its Clause-②: yes declaration no longer calls for a contract-review-tier review. The lane seat's own default-tier review, plus the gates (widening tells, pin tests, dispatch-gates --tier), is the review of record, and the build stays at the default tier. Unchanged: the Clause-② declaration itself, the manual floor for widenings under 代裁, and the routing rule that a diff touching packages/spec goes to the spec seat, where the contract-review-tier review still applies. This comment changes no label, assignee or claim.

Generated by Claude Code

Navigation Menu

Repository navigation

finding(types/components): `PageNodeSchema.body` is array-only while its only reader casts around that to accept a single node — and the root README's flagship example authors the refused form #8310

What

Measured (on `0ea7054f8`, through the doc-snippet gate's own program)

Why it matters

The decision, not made here

答异议(`pm:retriage`)—— 状态给 `needs-user-decision`,预挂的 `needs:contract-review` 同笔摘掉

一句话问题

选项 × 真实代价

业务含义直译

四棱(`os-decision-facets`)

裁后执行

维护者速读

Ruling recorded — A: `PageNodeSchema.body` widens back to `SchemaNode | SchemaNode[]`, matching `BaseSchema` and the runtime; the reader's cast goes (director seat, summon #17, decision batch #2, 2026-09-07)

1 remaining item

How it surfaced

⭐ One reading this seat adds, which strengthens the "which side is right" question

The three options, with a four-axis reading — ⛔ offered as input, ⛔ not a decision

What objectui#8125 did instead, so the record is clear

零、⛔ 本卡已有总监席裁决，⛔ 派发席不重裁、⛔ 你也不要重开

一、⛔⛔ 一个会让你做反的陷阱：字母在本卡上是冲突的

二、⚠️ 一条必须写进 PR、⛔ 不许被夸大的天花板

三、判别力

四、⛔ 边界

五、Clause-② 与载体

六、⭐ 位置轴与工具链

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(types/components): PageNodeSchema.body is array-only while its only reader casts around that to accept a single node — and the root README's flagship example authors the refused form #8310

Description

What

Measured (on 0ea7054f8, through the doc-snippet gate's own program)

Why it matters

The decision, not made here

Activity

os-zhuang commented on Sep 7, 2026

hotlong commented on Sep 7, 2026

os-zhuang commented on Sep 7, 2026

答异议(pm:retriage)—— 状态给 needs-user-decision,预挂的 needs:contract-review 同笔摘掉

一句话问题

选项 × 真实代价

业务含义直译

四棱(os-decision-facets)

裁后执行

维护者速读

hotlong commented on Sep 7, 2026

Ruling recorded — A: PageNodeSchema.body widens back to SchemaNode | SchemaNode[], matching BaseSchema and the runtime; the reader's cast goes (director seat, summon #17, decision batch #2, 2026-09-07)

1 remaining item

claude commented on Sep 8, 2026

How it surfaced

⭐ One reading this seat adds, which strengthens the "which side is right" question

The three options, with a four-axis reading — ⛔ offered as input, ⛔ not a decision

What objectui#8125 did instead, so the record is clear

os-zhuang commented on Sep 9, 2026

零、⛔ 本卡已有总监席裁决，⛔ 派发席不重裁、⛔ 你也不要重开

一、⛔⛔ 一个会让你做反的陷阱：字母在本卡上是冲突的

二、⚠️ 一条必须写进 PR、⛔ 不许被夸大的天花板

三、判别力

四、⛔ 边界

五、Clause-② 与载体

六、⭐ 位置轴与工具链

claude commented on Sep 10, 2026

huangyiirene commented on Sep 10, 2026

huangyiirene commented on Sep 10, 2026

claude commented on Sep 10, 2026

os-justin commented on Sep 10, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding(types/components): `PageNodeSchema.body` is array-only while its only reader casts around that to accept a single node — and the root README's flagship example authors the refused form #8310

Measured (on `0ea7054f8`, through the doc-snippet gate's own program)

答异议(`pm:retriage`)—— 状态给 `needs-user-decision`,预挂的 `needs:contract-review` 同笔摘掉

四棱(`os-decision-facets`)

Ruling recorded — A: `PageNodeSchema.body` widens back to `SchemaNode | SchemaNode[]`, matching `BaseSchema` and the runtime; the reader's cast goes (director seat, summon #17, decision batch #2, 2026-09-07)