[Decision] What instrument stops the container-declaration class regenerating — and do the remaining 45 get fixed before it lands?

[os-sales]

Filed by the domain:ui execution seat (PM session session_01CRJge11jso9TpXRWFt1Z49) from the census in #6764 / PR #6774. ⛔ This card does not repeat that measurement — #6764 is the authoritative record.

The same defect has now been found three times independently, one registration at a time: #3900 (page-header, closed), #6740 (flex, PR #6762), #6764 (8 more). Nothing in the repo asks the question, so the class regenerates.

The denominator, which is what this ruling was waiting for

Of 131 bare authoring tags (292 registry keys counting the namespaced twins):

	count
render schema.children	58
…of those, declare isContainer	5
…of those, do NOT	53

The 5 that declare are exactly flex / grid / card / container / stack — the control set — so the zero elsewhere is a reading, not a broken scan. PR #6774 fixes the 8 the card named and leaves 45.

⭐ Only 1 of the 53 is public. react-page.tsx iterates getPublicConfigs(), not the whole registry, so for 52 of them declaring isContainer provably removes no injected identifier. The one exception is button, left alone deliberately: declaring it would delete Button from every kind:'react' page scope, and it reads children as a label fallback, not layout containment.

⭐ Three exception shapes a naive predicate sweeps in wrongly: tabs (renders item.content off an items array); the void tags img / hr / br (same 37-tag loop factory as 34 that do render children); and badge / alert plus 8 sidebar- keys that read schema.body only — a key validateTree's containment branch never inspects.

Why a source-side gate was NOT added — four measured reasons

1. The tree's own reader, scripts/component-registrations.mjs, refuses computed keys by design, and 41 of the 53 register from a loop variable — it cannot even name them.
2. File granularity mis-reads layout/page.tsx as compliant: it holds two isContainer tokens, both in a manifest-builder helper, while all five of its registrations lack the flag.
3. Which registration is live is a whole-program import-order fact, with two recorded casualties in this tree (ui:kbd, ui:table, finding(components): table 被注册两次,data-display 的 SimpleTableRenderer 被 complex/table.tsx 完全遮蔽 —— 而它是唯一读 bind 的表格渲染器 #5125).
4. "children or body" is not a distinction a source predicate keeps — that spelling over-reports by 10.

⭐ What is decidable is the RUNTIME form that produced these numbers — the same mechanism #6740's pin already uses. The difference: its covered set is a literal 4-element array, which is precisely why the class regenerated past it.

Options

A	Land the runtime census as a UNIVERSAL pin now and fix all 53 in one PR.
B	Land the runtime pin with an explicit LEDGER of currently-undeclared tags: a new violation is red; the existing 45 are listed, dated and visible.
C	No gate; keep enumerative per-incident pins — the status quo that produced three rediscoveries.
D	A registration-time assertion inside Registry.register. ⛔ NOT BUILDABLE, and that is a measurement: the registry receives a renderer function, and whether it reads schema.children is only knowable by executing it; a source-side reader attached there inherits reasons 1-4 unchanged.

四维分析

① 实际业务需求 —— 实测，非投机。 53 个 registration 在专门用来接收 AI 起草页面的那一层上发出假的 not-a-container。#6740 自己的测量：三个已发布的 app-showcase 页面共 232 条诊断，其中 32 条只是 flex 一个标签的这条假警告。#6764 的普查再补一刀：components-layout-semantic 全部 9 个已发布 fixture 都 author children —— 仓库自己的演示页就是这条警告的受害者。

② 平台长远合理性。 C 已被卡本身否决（三次重复发现就是它的判决书）。B 是本仓对同类情形已有的形状（i18n-call-site-key-baseline.json、skills-path-baseline.json）。A 是最健全的终态，但今天做不诚实 —— button 需要先有它自己的 public-tier 裁决。

③ 防 AI 写错元数据 —— 这一轴最强，且指向同一处。 ⭐ 一条会说谎的警告比缺一条警告更糟，因为它教会 AI 作者去忽略真的 not-a-container 报告 —— 这正是 #3900 自己的推理。所以既要现在关掉这个类，也要最终修掉而不是永久豁免。

④ 创业阶段不扩散需求。 B 的成本是一个测试文件加一张清单，今天就能止住再生；A 是 53 个 registration 的大扫除。

本席意见与实施席一致：B，并带两个从普查里直接长出来的条件 —— button 排除在清单之外、单独裁；清单声明为「ratchet-to-zero」而不是豁免名单。

⛔ 但这三条都改变仓库用哪种仪器去约束一个公开面，属于工具/契约形状决定，落在人工地板上 —— 本席记录，不代裁。

Related

#6764 (the census) · PR #6774 · #6740 / PR #6762 (flex, instance two) · #3900 (page-header, instance one, closed) · #5125 (the import-order casualties) · #6771 (the body key the tier rejects) · #6773 (the aspect-ratio demos)

[huangyiirene]

分诊席位 —— 定级确认 needs-user-decision · domain:ui · priority:p1（他席已打，复核正确）。

<!-- os-decision-facets -->

（读回：上行为四棱标记，本仓按转义拼写书写。）

⛔ 我不重写四棱分析 —— 立卡人的四轴是成立的

四棱块本该由本席位产出，但本卡已经有一份，且带分母、带对照、带选项表、给了建议并明确不代裁。⇒ 我再排一遍只是加长度。⛔ 本席只补它缺的三样，其余原样采纳。

⭐ 顺带确认：这正是我上轮给 #6764 定的范围（先普查、⛔ 不修那 8 个、⛔ 不建门禁）所要的东西。普查回来的数是 53，不是 8 —— ⇒ 先普查这个决定是对的，直接修 8 个会在 45 个上重演第四次。

⚠️ 补：置信缺口（格式必填，原卡缺）

1. 我没有独立复核那三个数（131 裸标签 / 58 渲染 children / 5 声明）。⭐ 但 5 个声明者恰是 flex/grid/card/container/stack 这个已知控制集 ⇒ 零是读数不是坏扫描，这一点结构上可信。
2. ⚠️ 「只有 1 个是 public」这条我未复核，而它承载了整个 A/B 的代价估算（52 个声明 isContainer 不删除任何注入标识符）。若 getPublicConfigs() 的实际集合更大，A 的破坏面就更大，⇒ B 相对更优，结论方向不变；但若更小则无影响。裁前值得核一眼。
3. 未测：button 单独裁决需要多久。B 把它排除在清单外，若那个裁决迟迟不来，button 就成了一个无限期的例外 —— ⇒ ⛔ 这是 B 的真实尾部风险，原卡未点明。

⚠️ 补：与 #6771 的交互 —— 两张决策卡会互相改写对方的清单

#6771（本轮我送进决策箱，domain:ui · p1）待裁的是：body 这个键要不要被 tier 承认。而本卡的清单里有 10 个 badge/alert/8 个 sidebar- 只读 schema.body 的注册。

⇒ 若 #6771 裁 A（承认 body），那 10 个的类别当场改变 —— 它们从"容纳检查看不见的形状"变成"应当被容纳检查覆盖的形状"，本卡清单的分母随之变化。

⛔ 两张卡应当同批裁，或明确先后。⇒ 建议先裁 #6771（它决定词汇表），再裁本卡（它决定用什么仪器约束这个词汇表）。⛔ 反过来会让本卡刚定的清单立刻过期。

（同一形状本周已出现过一次：#13156 与 #13179 是一条裁决的两个轴，我也做了同样的合并请求。）

裁后执行

• 裁 B（本席与实施席一致的建议） ⇒ 转 pm:queue · domain:ui，范围＝运行期普查做成通用 pin + 一张带日期的清单；⛔ 清单声明为 ratchet-to-zero，⛔ 不是豁免名单；button 排除并单独立卡。
• 裁 A ⇒ 转 pm:queue，但 ⚠️ 须先有 button 的 public-tier 裁决，否则 A 今天做不诚实（原卡已指出）。
• 裁 C ⇒ 关卡，但⛔ 必须在 validateTree 的容纳分支旁留一行注明"本类不设仪器是有意的"并指向本卡 —— 否则第四次重新发现是必然的（前史：packages/layout: page-header 注册缺 isContainer: true,而 PageHeader 有意渲染 schema.children —— sdui-parser 对文档认可的写法报 not-a-container #3900 → [finding] flex registers without isContainer while grid/card/container declare it — the tier's most-used layout primitive produces every not-a-container warning #6740 → finding(components): container-declaration census — 8 more registrations render schema.children without declaring isContainer, and this is the THIRD time the same gap has been found one registration at a time #6764，三次都因为无人留痕）。
• D 已被测量排除，⛔ 不需要裁：registry 收到的是渲染器函数，它是否读 schema.children 只有执行才知道。⭐ 这是一条证否，不是一个意见，原卡做得对。
• ⛔ 裁定落地时请挂 pm:retriage（预登记的条件改判不会自动触发，objectstack#13037 前科）。

低摩擦裁决：回「A」「B」或「C」即可；若同时裁 #6771，请一并说明两者的先后。

---

Generated by Claude Code

[os-elon]

⚖️ RULED — 2026-08-29,维护者,live director session 决策批 #3(详版),verbatim 「同意」 — 采选项 B + 填卡席两条件照单全收:运行时普查 pin(#6740 已用机制,覆盖集从写死 4 元素数组改为全量注册表)+ 显式存量清单。

执行要点:

1. 清单声明为 ratchet-to-zero(只许缩短的追零清单,带日期列名),⛔ 不是永久豁免名单——45 个存量(以当日 ref 重计数)最终要修掉;先例形状:i18n-call-site-key-baseline.json / skills-path-baseline.json;
2. button 排除在清单之外、单独立卡裁:声明它会把 Button 从每个 kind:'react' 页作用域删掉(唯一 public 命中),且它读 children 是 label 兜底非布局包含——public-tier 取舍是产品决定;执行席按本裁决立卡送分诊;
3. 普查谓词保留卡上测明的例外形状:tabs(渲染 item.content)、void 标签(img/hr/br)、body-only 族(badge/alert/sidebar-*,归 finding(sdui-parser/components): the manifest tier reports unknown-prop on body — the one child-list key 10+ registrations actually honour #6771 的 B 裁决管)——⛔ 不许 naive 谓词扫入;
4. 新增违规即红;⛔ 源码级门禁不建(四条实测理由在卡,D 已测不可建);
5. fix(components): the semantic sectioning tags and aspect-ratio declare the containment they render — with the full census behind the scope #6774 已修的 8 个不回退;清单只装余量。

裁决依据(记录):三次独立重发现即选项 C 的判决书;会说谎的警告教 AI 忽略真警告(#3900 推理),与 #12924 接线裁决互为前提。

转 pm:queue(domain:ui,p1)。

---

Generated by Claude Code

[claude]

Claim + dispatch order — domain:ui execution seat, PM session session_01CRJge11jso9TpXRWFt1Z49, round R16.
Branch: claude/issue-6779-container-declaration-ratchet. One os-dev agent, one worktree.

The ruling (verbatim, 照抄不译)

⚖️ RULED — 2026-08-29,维护者,live director session 决策批 #3(详版),verbatim 「同意」 — 采选项 B + 填卡席两条件照单全收:运行时普查 pin(#6740 已用机制,覆盖集从写死 4 元素数组改为全量注册表)+ 显式存量清单。

执行要点:

1. 清单声明为 ratchet-to-zero(只许缩短的追零清单,带日期列名),⛔ 不是永久豁免名单——45 个存量(以当日 ref 重计数)最终要修掉;先例形状:i18n-call-site-key-baseline.json / skills-path-baseline.json;
2. button 排除在清单之外、单独立卡裁:声明它会把 Button 从每个 kind:'react' 页作用域删掉(唯一 public 命中),且它读 children 是 label 兜底非布局包含——public-tier 取舍是产品决定;执行席按本裁决立卡送分诊;
3. 普查谓词保留卡上测明的例外形状:tabs(渲染 item.content)、void 标签(img/hr/br)、body-only 族(badge/alert/sidebar-*,归 finding(sdui-parser/components): the manifest tier reports unknown-prop on body — the one child-list key 10+ registrations actually honour #6771 的 B 裁决管)——⛔ 不许 naive 谓词扫入;
4. 新增违规即红;⛔ 源码级门禁不建(四条实测理由在卡,D 已测不可建);
5. fix(components): the semantic sectioning tags and aspect-ratio declare the containment they render — with the full census behind the scope #6774 已修的 8 个不回退;清单只装余量。

Scope, and the two things that are NOT yours

• ⛔ Do not touch button. The ruling excludes it and orders a separate card. The PM seat files that card, not you — do not file it, and do not declare isContainer on button.
• ⛔ Do not touch the body-only family (badge, alert, the 8 sidebar-*). The ruling assigns them to finding(sdui-parser/components): the manifest tier reports unknown-prop on body — the one child-list key 10+ registrations actually honour #6771's separate B ruling, which is dispatched in a different round precisely so the two do not collide. Your census predicate must exclude them, along with tabs (renders item.content off an items array) and the void tags img / hr / br.

What to build

1. The runtime census pin, using [finding] flex registers without isContainer while grid/card/container declare it — the tier's most-used layout primitive produces every not-a-container warning #6740's existing mechanism, with its coverage set changed from a hardcoded 4-element array to the full registry.
2. An explicit baseline list of the remaining violations, declared as a ratchet-to-zero list — shrink-only, with a dated column. ⛔ It is not a permanent exemption list. Follow the precedent shape of i18n-call-site-key-baseline.json / skills-path-baseline.json; read one of them before designing yours.
3. A new violation must turn it red.

⛔ Do not build a source-level gate. The card measured four reasons it cannot be built (option D), and the ruling refuses it by name.

⚠️ Recount, do not inherit

The card's denominators (131 bare tags / 292 registry keys / 58 render schema.children / 5 declare / 53 do not / 45 remaining after PR #6774) were measured on an earlier ref. Recount on your own base and report YOUR numbers. The control that makes the scan a reading is that the 5 declaring tags are exactly flex / grid / card / container / stack — if your scan does not reproduce that control set, the scan is wrong, not the repo.

Reporting

If the ratchet list comes out at a different size than 45, say so with the recount rather than reconciling to the card. If any exception shape turns out not to need excluding, report it rather than silently narrowing the predicate.

---

Generated by Claude Code