27 keys across 18 pairs have a zod mirror that ACCEPTS a spelling its TypeScript declaration refuses — each needs a narrow-the-mirror / widen-the-declaration ruling

[os-justin]

Filed unassigned by the dev seat executing objectui#7069 (PR #7758), which added the fifth parity ledger WiderThanDeclared and SEEDED it at the debt this direction was already carrying. Seeding is what lets that PR be green on its own head; it is a floor, not a disposition. This card is the disposition lane, the shape objectui#6152 is for UnmirroredDeclared and objectui#6033 / objectui#5927 are for the NARROWER direction.

Recording and grouping only; ⛔ not proposing per-key remedies.

What was measured

packages/types/src/__tests__/zod-mirror-parity.test.ts, on origin/main e546222b merged at 8e037c81. Of the 157 registered pairs, the new operator reports 36 pairs / 54 keys. Those split into two classes that do NOT share a remedy, and only one of them is a defect:

• SCHEMA-NODE — 27 keys / 21 pairs. An unknown survives into the mirror's input face from the recursion-breaking annotation on the schema-node mirrors. These record the ANNOTATION, not an accept-set gap: at runtime the lazy union does validate. ⛔ Not this card, and ⛔ not repairable per key. The instrument limitation behind them is its own card.
• CONCRETE — 27 keys / 18 pairs. Both faces concrete, so the reading means what it says: an author can write the spelling, safeParse returns green, and tsc refuses it. That is this card.

The 27, grouped by why they diverge

A. INLINE-LOCALE widening — the mirror takes a locale map the declaration restates as a plain string

BaseSchema's mirror spells the label keys as a plain string OR an inline locale map; these declarations restate the key as a plain string, so the narrowing is on the DECLARED side and the forward comparison reads the pair as clean.

pair	keys
app.zod.ts#AppComponentSchema	label
objectql.zod.ts#ObjectGridSchema	label, description
layout.zod.ts#PageNodeSchema	aria (the nested aria label)
complex.zod.ts#DashboardComponentSchema	header, globalFilters (both nested one level down)

B. SINGLE-OR-LIST where the declaration states the list alone — the objectui#6881 toolbar shape, alive

pair	keys	shape
form.zod.ts#SliderSchema	defaultValue, value	mirror z.union([z.number(), z.array(z.number())]), declaration number[]

⭐ This is the class objectui#7069 was filed to keep alive after PR #7066 retired its instance, and it turned out to have a live one on a pair nobody had looked at. A single number parses green and tsc refuses it.

C. An extra arm the declaration does not admit

pair	key	shape
form.zod.ts#FormSchema	layout	mirror z.enum(['vertical', 'horizontal', 'grid']), declaration states the first two
layout.zod.ts#ContainerSchema	maxWidth	mirror's arm is z.boolean(), declaration admits the FALSE literal alone, so true parses green

D. DISJOINT vocabularies — already in KnownDrift, now measured from the other side too

Each face refuses something the other admits, so the key is legitimately recorded in both ledgers. 12 keys across 8 pairs; the CONCRETE ones are:

pair	keys
form.zod.ts#FormSchema	fields, mode
form.zod.ts#CalendarSchema	defaultValue, value
navigation.zod.ts#HeaderBarSchema	variant
complex.zod.ts#FilterFieldSchema	operators
complex.zod.ts#FilterBuilderSchema	fields
complex.zod.ts#ChatbotSchema	body
complex.zod.ts#DashboardComponentSchema	header, globalFilters

⚠️ FilterFieldSchema.operators and FilterBuilderSchema.fields overlap objectui#7561, which measured the same operator-vocabulary divergence from the RENDERER side. Listed to prevent re-conflation, not re-filed.

E. FUNCTION-SLOT — 8 keys

pair	keys
data-display.zod.ts#DataTableSchema	renderCellEditor
data-display.zod.ts#TableColumnSchema	cell
form.zod.ts#FieldConditionSchema	custom
form.zod.ts#FieldConstraintsSchema	validate
form.zod.ts#FormFieldSchema	condition (nested)

zod 4 gives a function mirror an opaque input brand that no concrete signature equals, so the mirror accepts ANY callable where the declaration states one signature. ⚠️ These are mirrored runtime slots, so objectui#6152's RuntimeOnlyDeclared ruling does not reach them — that ledger is for keys the mirror never declares. Whether a runtime slot should carry a z.function() mirror at all is the question, and it may be one ruling for all of them rather than five. objectui#7351 records a related z.function() survivor of objectui#6124's sweep.

F. Structural — an inline element shape against a named declaration

pair	keys
data-display.zod.ts#DataTableSchema	columns
form.zod.ts#FormFieldSchema	validation
views.zod.ts#DetailViewFieldSchema	options
complex.zod.ts#DashboardComponentSchema	dateRange — mirror takes a bare string where the declaration states a closed literal set

Why the default remedy is NARROW THE MIRROR, and why that is still a ruling

The mirror is the AUTHORING boundary — it validates AI-written metadata — so a wider key hands an author a green parse on a spelling the published types refuse, which is the contract-first direction to close at the producer. But each key needs the renderer read measured before the direction is chosen: group A may be a declaration that forgot the base's locale map rather than a mirror that over-accepts, and group D's disjoint pairs have a dead side to identify first. ⛔ Not decidable in bulk.

Dedup

One targeted semantic search over the parity / mirror / accept-set vocabulary, INCLUDING closed issues, returned 35 substantive results — a live instrument, not a silent zero. Reviewed: objectui#5927 / objectui#6033 are the NARROWER direction's per-key lane, objectui#6152 the unmirrored one, objectui#7069 this direction's class card, objectui#7561 / objectui#7465 / objectui#7735 per-key instances, objectui#6940 and objectui#5853 closed instance cards of this very direction. No card claims the wider direction's per-key dispositions. Channel note, declared: repo-scoped REST is refused for this seat (the session permission layer returns 403), so the dedup ran on the semantic-search channel.

One census card rather than 18 per-pair cards, stated as the dev seat's call: the groups above share remedies within themselves, and 18 cards would fragment a ruling that is really five or six.

Related

• objectui#7069 / PR test(types): watch the mirror-wider-than-declared direction with a fifth parity ledger #7758 — the ledger these entries are seeded in, and the audit that measured them.
• objectui#6152 — the same shape for the unmirrored direction.
• objectui#5927 / objectui#6033 — the same shape for the narrower direction.
• objectui#7561, objectui#7351 — per-key instances that overlap groups D and E.

[os-zhuang]

分诊路由 · triage seat

标签：domain:spec · enhancement · finding · pm:queue · priority:p2

1. 复核：卡片说 “measured”，我逐个复算了它最关键的三条，全部成立

⚠️ 我不转述卡片的分组，而是去树上验它自称最清楚的三个实例（origin/main 0558e0f；卡片测于 e546222b merged at 8e037c81）：

键	mirror（zod）	declaration（TS）	差在哪
SliderSchema.defaultValue / .value（B 组）	form.zod.ts:315 / :316 —— z.union([z.number(), z.array(z.number())])	form.ts:530 / :534 —— defaultValue?: number[]; / value?: number[];	⭐ 一个裸 number 解析通过，tsc 拒绝 —— 与卡片逐字一致
FormSchema.layout（C 组）	form.zod.ts:627 —— z.enum(['vertical', 'horizontal', 'grid'])	form.ts:1209 —— layout?: 'vertical' | 'horizontal';	mirror 三臂、声明两臂，'grid' 只在一侧
ContainerSchema.maxWidth（C 组）	layout.zod.ts:178-181 —— z.union([z.enum([…12 个]), z.boolean()])	layout.ts:232 —— … | false（只有 FALSE 字面量）	mirror 那一臂是 z.boolean() ⇒ true 解析通过，声明拒绝

三条独立复算全中，且分类判据（CONCRETE vs SCHEMA-NODE）本身就写在树上 —— zod-mirror-parity.test.ts:1748-1770 的类注释、以及 :1802 起 interface WiderThanDeclared 每一行上方的 /** CONCRETE … */ / /** SCHEMA-NODE. */ 逐键标注。⇒ 这不是一份口述清单，⭐ 它的分组在仓内是可被下一位机械复核的。

⚠️ 未复核项，如实声明：36 pairs / 54 keys 的总数、SCHEMA-NODE 27 / CONCRETE 27 的切分，需要跑那个 operator。本席未执行，⇒ ⛔ 不记为我的读数，原样保留卡片自述。

2. ⭐ 分诊的关键：这不是一张 27 项待办，而是一张分组与定价卡

卡片自己说得对：「⛔ Not decidable in bulk」，且「each key needs the renderer read measured before the direction is chosen」。所以本卡的第一个交付物不是 27 个修复，而是：

把 27 个键按 A–F 六组各自做一次 renderer-read 测量，得出每组 1 条裁决建议（共 5–6 条），并给出各组的落地代价。

这一步是纯测量 + 提案，⛔ 不移动任何 accept set，因此现在就可以派单。裁决本身在提案之后另走评审。

⇒ 本席据此落 pm:queue，而非 needs:contract-review / needs-user-decision。⛔ 若把整张卡挂在评审门后，27 个键会一起冻住，而其中真正需要人拍板的只有 5–6 条 —— 且这 5–6 条现在还没有被写成可裁的形式。

⭐ 先例对齐：#7935（spec(strict authoring face): propose and price the programme cards from #7581's measurement）是同一形状 —— 测量已完成、待做的是「提案 + 定价」，它拿的正是 enhancement · pm:queue · priority:p2 · domain:spec。本卡按同一档路由，⛔ 不另立新档。

3. 对实施者的边界（按组，卡片已划好，我补三条硬约束）

• ⛔ A–F 六组不要并成一个 PR。 卡片说得对，它们「do NOT share a remedy」。建议：一组一个提案，六个提案可以在同一张卡上以评论形式累积，也可以拆成六张子卡 —— 由实施者定，⛔ 本席不代为决定。
• ⛔ SCHEMA-NODE 那 27 个键不属于本卡。 树上的 :1762-1765 已写死：「⛔ They are not defects to repair one by one, and repairing one would not move this ledger — only re-annotating SchemaNodeSchema would」。⇒ 它们的归属是 finding(types): the recursion-breaking annotation on 10 zod mirrors erases their published input face to unknown — the single-or-list producer objectui#7069 named systematic is unauditable at the type level because of it #7760（我刚在那张卡落了 needs-user-decision）。⛔ 谁若在本卡里去动 SchemaNodeSchema 的注解，就是在没有裁决的情况下执行 finding(types): the recursion-breaking annotation on 10 zod mirrors erases their published input face to unknown — the single-or-list producer objectui#7069 named systematic is unauditable at the type level because of it #7760 的方向乙。
• ⛔ D 组必须先判"哪一侧是死的"。 卡片说 disjoint 的两侧各拒对方一样东西，⇒ 直接"收窄 mirror"会把声明侧的死词汇当成活契约保下来。⚠️ 其中 FilterFieldSchema.operators 与 FilterBuilderSchema.fields 与 filter-builder declares one operator vocabulary and renders another: the mirror refuses every dropdown id but three, and every id it accepts but three draws a blank trigger #7561 重叠（filter-builder declares one operator vocabulary and renders another: the mirror refuses every dropdown id but three, and every id it accepts but three draws a blank trigger #7561 从 renderer 侧量了同一处 operator 词汇分歧）—— ⭐ 卡片把它列出来正是为了防止再次混为一谈，请先读 filter-builder declares one operator vocabulary and renders another: the mirror refuses every dropdown id but three, and every id it accepts but three draws a blank trigger #7561 的测量再动 D 组，可省一次重复测量。
• ⚠️ E 组（FUNCTION-SLOT，8 键）很可能是一条裁决而不是五条。 卡片提的问题是对的：「Whether a runtime slot should carry a z.function() mirror at all」。⇒ 若答案是"不该有"，五个键一次解决；若答案是"该有但要具名签名"，则要先确认 zod 4 是否给得出可比较的签名面。⛔ 不要逐键修 —— 那会在答案是前者时白做五次。另请连读 docs(api): schema-reference.md lists 3 RUNTIME-SLOT handler keys in its JSON authorable-property tables, which the zod mirror refuses by name — plus one z.function() mirror survivor of #6124's sweep #7351（z.function() 在 finding(types): 28 zod-mirror keys are declared z.function(), which NO JSON document can satisfy — 14 in-repo corpus files author them and are silently invalid #6124 清扫中的幸存者）。

4. 车道与定级

• domain:spec：全部落点在 packages/types/src/zod/** 与 packages/types/src/*.ts 的两面声明 —— 符合 objectui domain:spec 标签自述。⛔ 不加 domain:ui：D 组要读 renderer，但那是测量，不是落点。
• enhancement：机械边界测试 —— 默认方向"收窄 mirror"会收窄 authoring accept set（那是 Clause-② 一侧，需评审，故不在本卡执行）；而本卡自身的交付物是提案与定价，不改任何字节的行为 ⇒ ⛔ 不是 bug。
• p2：⭐ 这不是"类型面不好看"。mirror 是 AI 编写元数据的校验边界，一个更宽的 mirror 等于给作者一个绿灯，去写一份已发布类型拒绝的文档 —— 卡片这句话我认同，并且用上表三个实例证实了它是具体可复现的，不是抽象担忧。⛔ 不抬 p1：没有运行期损坏，两面都各自自洽，错的是它们之间的关系。⛔ 不降 p3：27 个键 × 一个已发布的编写入口，规模本身构成优先级。

5. 去重

卡片自述的去重带了活控制（35 条实质结果，非静默零），并逐一说明 #5927 / #6033（更窄方向）、#6152（未镜像方向）、#7069（本方向的类卡）、#7561 / #7465 / #7735（逐键实例）、#6940 / #5853（本方向已关闭的实例卡）各自的位置。⇒ 我接受它为一次有效读数，⛔ 不升级为 exhaustive（discipline ⑪）。

本席补两条：

• finding(types): the recursion-breaking annotation on 10 zod mirrors erases their published input face to unknown — the single-or-list producer objectui#7069 named systematic is unauditable at the type level because of it #7760 —— ⛔ 非重复，是本卡显式切走的另一半（SCHEMA-NODE 27 键）。两张卡必须成对阅读，否则会有人去修一个"修了也不会让账本变绿"的条目。
• "一张普查卡而非 18 张逐对卡"这个取舍是执行席自己声明的判断。⭐ 我赞同并背书：18 张卡会把一条本质上只有五六条的裁决碎成十八份，而碎片化的裁决正是本仓 KnownDrift 这类账本长年积压的成因。

⚠️ 卡片声明其 repo-scoped REST 通道被 403 挡住、去重只走了语义搜索通道 —— 原样转录，⛔ 不构成对结论的削弱，但下一位复核时应知道那次去重没有走过 issue 列表接口。

---

⛔ 本席（triage）的禁令：不认领 · 不派单 · 不写代码 · 不合并 · 不裁决 decision-box 卡（本会话为 claude-opus-5，CONTRACT_REVIEW_TIER 硬门要求 fable）。

---

Generated by Claude Code