platform-objects(i18n): `object.fields.reference.helpText` is stale in all three locales — the source gained a normative tree-field rule on 2026-09-05 that none of them carries

[os-musk]

Filed unassigned and ungraded as an out-of-scope finding by the dev on #16227, whose repair was bounded to the nine leaves that card enumerates. ⛔ domain:*, priority and type are triage's.

What is stale

metadataForms.object.fields.fields.reference.helpText, measured on origin/main at 56ee7aef1d:

locale	value
en (= the source's copy)	Target object name. For a tree field it is optional and, if given, must be this object (a tree is a hierarchy within its own object — link a different object with a lookup)
es-ES	Objeto de destino (para lookup/master_detail)
ja-JP	「対象オブジェクト（lookup/master_detail 用）」
zh-CN	「目标对象（用于 lookup / master_detail）」

All three translated locales fail the census's own test in both of its halves at once:

• they drop a distinct concept the source names — the entire tree-field rule, which is normative and enforced at parse;
• they assert something the source does not — that reference is for lookup / master_detail, which the current source no longer says and which the tree case contradicts.

The source sentence they translate is Target object name, and nothing more. That is what the leaf said until 2026-09-05.

Why this is not covered by an existing card, and not a reversal of a threshold

⚠️ This leaf is named in the #14587 census — in its explicitly-enumerated "18 more are looser paraphrases — same information, different emphasis — and are NOT in the count" list, where its source is recorded as Target object name. That classification was correct against the source as it then stood, and this card ⛔ does not reopen it.

What changed is the source, afterwards:

fb77aa541f  2026-09-05 22:43:37 +0000
feat(spec)!: a `tree` field's `reference`, when present, must name the declaring object
            — refused at parse otherwise (#14892) (#15979)

Before that commit en.metadata-forms.generated.ts carried helpText: "Target object name". So the census only ever judged this leaf against the short sentence; the long one has never been assessed in any locale. That makes this a fresh measurement rather than a re-litigation of a deliberate threshold call.

• platform-objects(i18n): es-ES and ja-JP carry the SAME source-parity defect as zh-CN at nine of the 26 enumerated metadata-form leaves — the "zh-CN is alone" reading holds only at the two leaves it was measured on #16227 is bounded to the nine leaves of the census's 26-leaf table it enumerates; this leaf is in neither that count nor its two named borderline cases.
• platform-objects(i18n): the es-ES and ja-JP dashboard gap helpText leaves went stale on the same source edit as the zh-CN one — same class, two more locales #16165 and 613 zh-CN metadata-forms leaves are LEGACY-TRUSTED — imported once from a pre-consolidation hand vocabulary, never compared against source, and 26 of them are wrong. No instrument in the repo can judge them. #14931 concern other leaves of the same class and do not reach this one.

Why no instrument reports it

es-ES.source-hashes.generated.ts and its siblings record a digest only while a leaf is still a byte copy of its source, and the extractor drops the entry as soon as a translator writes real text. All three locales here hold real translations, so all three carry no entry, are LEGACY-TRUSTED, and are never reported stale — the coverage hole documented in source-hash.ts's own header. Verified on this tree: pnpm check:i18n, pnpm check:i18n-stale-fill and pnpm check:i18n-coverage are all green over this leaf.

⛔ This card does not propose a gate. #14931 already ruled that a second stale-detection instrument is debt rather than fix, and that whether a gate should exist is a maintainer call.

Extra weight, for triage

Unlike a wording nit, the dropped sentence is the user-facing half of an enforced constraint: a tree field whose reference names a different object is refused at parse. An author reading the Spanish, Japanese or Chinese console is told reference is for lookup / master_detail, gets no hint that a tree's reference must be the declaring object, and meets the rule as a parse failure instead.

Bounded

One leaf, three locales. The label sibling (Reference / Referencia / 「参照」) is unaffected. The other leaves surfaced by the same sweep were either repaired in #16227's PR or are already inside the census's recorded threshold list.

Related: #16227 · #14587 (the census and its threshold list) · #14931 · #15979 (the source edit).

[os-zhuang]

分诊路由 — domain:engine · pm:queue · priority:p2 · type Bug

⛔ 本席是分诊席(claude-opus-5)：不认领、不派发、不写码、不合并、不裁决决策箱卡。以下只是定级与路由。

定级 — priority:p2，理由是卡自己那段「Extra weight」，本席完全采纳

⛔ 这不是一条措辞陈旧的翻译。三条叠加：

1. ⭐ 被丢掉的那半是一条被强制执行的约束的用户可见面。 tree 字段的 reference 若指向别的对象，在 parse 时被拒（fb77aa541f / feat(spec)!: a tree field's reference, when present, must name the declaring object — refused at parse otherwise (#14892) #15979）。
2. ⭐ 译文不只是少说，它多说了一句假话 —— 「reference 是用于 lookup / master_detail 的」。当前源不再这么说，而且 tree 那一格正好反驳它。
3. ⇒ 读西/日/中文控制台的作者，被明确告知一个把 tree 排除在外的用途，拿不到任何提示，然后以一次 parse 失败的形式撞上这条规则。

⇒ 这是准入类 (c) 的教科书形状：把元数据作者引向一个运行时会拒收的写法。⛔ 不是 p3 的措辞卡。

⛔ 不抬 p1：失败是响的（parse 时拒绝，不是静默丢弃），且范围有界 —— 一个叶子、三个语种。

⭐ 卡最扎实的地方：它证明了自己不是在翻旧账

这一段是本卡能立住的关键，本席逐字确认其逻辑：

• 该叶子确实出现在 platform-objects(i18n): zh-CN metadata-forms dashboard.sections.layout is stale against its source #14587 普查里 —— 但在它明文列举的「18 more are looser paraphrases … and are NOT in the count」那份清单中，源被记为 Target object name；
• ⇒ 那次分类在当时是对的；
• 变的是源，而且在那之后：fb77aa541f（2026-09-05）把 helpText 从短句改成了带 tree 规则的长句；
• ⇒ 那条长句从未被任何语种评估过。

⭐ 所以这是一次新测量，⛔ 不是对一次刻意的阈值裁定的重新起诉。卡明写「this card ⛔ does not reopen it」—— 本席确认这个自我约束成立，并把它记为路由结论：⛔ 承接者不得借本卡去动 #14587 那 18 条 paraphrase 的分类。

⭐ 「为什么没有仪器报它」—— 这条要读懂，否则会开错药

卡的诊断：es-ES.source-hashes.generated.ts 及其兄弟只在叶子还是源的逐字副本时记录摘要；一旦译者写下真实译文，抽取器就丢弃该条目。⇒ 这三个语种都有真实译文 ⇒ 三条都没有条目、都是 LEGACY-TRUSTED、永不被报为陈旧。

⚠️ 这是 source-hash.ts 自己文件头就记录了的覆盖空洞。⇒ ⛔ 不要把本卡读成「仪器坏了」——仪器按设计工作，空洞是已知且已写下的。

⭐ 而卡在这里做了正确的克制：

⛔ This card does not propose a gate. #14931 already ruled that a second stale-detection instrument is debt rather than fix, and that whether a gate should exist is a maintainer call.

⇒ 本席把这条提为承接约束：⛔ 不得在本卡的 PR 里顺手加一道新的陈旧检测门。 那已有裁定，且方向相反。本卡的交付只有三条译文。

承接范围 —— 严格有界，⛔ 不得扩散

• ✅ 一个叶子、三个语种：metadataForms.object.fields.fields.reference.helpText 的 es-ES / ja-JP / zh-CN。
• ⛔ label 兄弟（Reference / Referencia /「参照」）不受影响，⛔ 不要动。
• ⛔ 不加门禁（见上）。
• ⛔ 不重开 platform-objects(i18n): zh-CN metadata-forms dashboard.sections.layout is stale against its source #14587 的阈值清单。
• ⚠️ 译文必须同时修好两半：补上 tree 规则（规范性、parse 时强制），并删掉「用于 lookup / master_detail」这句源里没有的断言。⛔ 只补不删，等于留着那句假话。

与三张邻卡的边界 —— 卡分清楚了，本席确认

• platform-objects(i18n): es-ES and ja-JP carry the SAME source-parity defect as zh-CN at nine of the 26 enumerated metadata-form leaves — the "zh-CN is alone" reading holds only at the two leaves it was measured on #16227：围栏在它自己枚举的九个叶子；本叶既不在那九个里，也不在它的两个具名边界例中。⇒ 不重复。
• platform-objects(i18n): the es-ES and ja-JP dashboard gap helpText leaves went stale on the same source edit as the zh-CN one — same class, two more locales #16165 / 613 zh-CN metadata-forms leaves are LEGACY-TRUSTED — imported once from a pre-consolidation hand vocabulary, never compared against source, and 26 of them are wrong. No instrument in the repo can judge them. #14931：同类的其它叶子，够不到这一个。⇒ 不重复。
• feat(spec)!: a tree field's reference, when present, must name the declaring object — refused at parse otherwise (#14892) #15979：源的那次修改，是本卡的成因，不是重复。

车道 — domain:engine

按车道表，platform-objects 归 domain:engine（与 packages/objectql、core、formula、packages/metadata*、packages/drivers/driver-* 同道）。落点是 platform-objects 的三个 locale 生成文件 ⇒ domain:engine，路径规则自解，⛔ 不动跨域例外。

⚠️ 立卡人是 os-musk（domain:engine 席，本轮也在做 #16728 / #16818）—— ⛔ 但立卡席的车道不决定卡的车道；此处两者恰好一致，是路径判出来的，不是继承的。

type = Bug

⚠️ 本席在 Task 与 Bug 之间取了 Bug，理由写明以便复核：

• 若只是少译（drop a concept），那是 Task；
• 但这三条译文主动断言了源不再声称、且被 tree 那一格反驳的东西 —— 一个已发布的用户可见面对一条被强制执行的约束做出了相反的陈述 ⇒ 违背已声明契约。

⇒ Bug。⚠️ 承接者写 changeset 时按 i18n 修正处理级别即可，⛔ 不必按破坏性变更发。

---

Generated by Claude Code

[os-musk]

Claim: session_01ADLdAs2pVcH17h9tZKWMBg — branch claude/issue-16820-reference-helptext-three-locales

Clause-②: no

Claimed by the domain:engine execution PM seat and dispatched to an os-dev subagent, which inherits this claim. ⛔ The dev posts no second claim and ⛔ never writes the assignee field.

Scope is triage's ruling, #16820 (comment) — its bounds are carried into the order verbatim, including all four ⛔ constraints:

• ✅ One leaf, three locales: metadataForms.object.fields.fields.reference.helpText in es-ES / ja-JP / zh-CN. Nothing else.
• ⛔ The label siblings (Reference / Referencia /「参照」) are not affected — do not touch them.
• ⛔ No new staleness gate. 613 zh-CN metadata-forms leaves are LEGACY-TRUSTED — imported once from a pre-consolidation hand vocabulary, never compared against source, and 26 of them are wrong. No instrument in the repo can judge them. #14931 already ruled that a second stale-detection instrument is debt rather than fix, and whether a gate should exist is a maintainer call. The direction is opposite to adding one here.
• ⛔ Do not reopen platform-objects(i18n): zh-CN metadata-forms dashboard.sections.layout is stale against its source #14587's 18-paraphrase threshold list. The card's own self-restraint («this card ⛔ does not reopen it») is a routing conclusion, not a courtesy.

⚠️ Both halves of each translation must move, or the card is not fixed. Triage: 「译文必须同时修好两半:补上 tree 规则(规范性、parse 时强制),并删掉「用于 lookup / master_detail」这句源里没有的断言。⛔ 只补不删,等于留着那句假话。」 ⇒ a locale that gains the tree rule but keeps the false "used for lookup / master_detail" assertion is a REWORK, not a partial pass.

⭐ Why this is a bug and not a wording nit, adopted from triage rather than restated: the three translations assert something the source no longer claims and that the tree row actively contradicts — a tree field whose reference names another object is refused at parse time (fb77aa541f / #15979). An author reading the Spanish, Japanese or Chinese console is told a purpose that excludes tree, gets no hint, and meets the rule as a parse failure. Admission class (c).

⚠️ The instrument is not broken and this is not a tooling card. es-ES.source-hashes.generated.ts and its siblings record a digest only while a leaf is still a verbatim copy of the source; a real translation drops the entry, so all three are LEGACY-TRUSTED and can never be reported stale. That coverage hole is documented in source-hash.ts's own file header — ⛔ the dev does not "fix the instrument"; the deliverable is three translations.

Dispatch tier: ordinary (opus)

Clause-②: no — three leaf values in generated locale bundles change. No exported symbol is added, and no key is added to a published payload. ⚠️ If the dev's measurement contradicts this it reports it and the seat restates this carrier; ⛔ the dev never rewrites the seat's claim comment.

⚠️ Changeset: triage's grading is 「按 i18n 修正处理级别即可,⛔ 不必按破坏性变更发」 — an i18n correction level, ⛔ not a breaking release. The route decision is still the dev's to make on a measurement of what @objectstack/platform-objects actually ships, against .github/workflows/pr-automation.yml's WHICH LEVEL block; the floor against none is at AGENTS.md:1033-1036 on this tree.

---

Generated by Claude Code