`PageTranslation.components` has no key for `element:text`'s `content` — the one string that element renders cannot be addressed by a bundle

[os-warren]

Found while authoring a read-only record page in an application (objectstack-ai/duly#13) with a complete zh-CN bundle, on @objectstack/spec 17.2.0.

The gap

PageTranslation.components (packages/spec/src/system/translation.zod.ts) declares the per-component key face as title | description | label | placeholder | emptyText | submitLabel.

element:text renders exactly one authored string, content, and it is not in that list. So a page that puts a sentence on screen through the element the platform provides for putting a sentence on screen has no bundle address for it.

Why this reads as a missed key rather than a decision

The schema's own comment says the face was derived from the copy props components declare, and it names its two deliberate exclusions — help and subtitle — with reasons. content is neither named nor excluded; it is simply absent. Every other authored string on the same page has a key.

What an application does today

The blessed alternative works: an inline I18nLabel map ({ en, 'zh-CN' }) on content, which is what the platform's own sys-user.page.ts does. The reporting app uses it, and its coverage walk records those strings under an untranslatable ledger with this reason, pinned by count (16 strings on one page). That is a workaround with a cost: those strings are invisible to os i18n extract, to check:i18n-coverage, and to a translator working from the bundle, and each one is an inline map in a page definition instead of a row in the locale file.

Suggested shape

Add content to PageTranslation.components, resolved in translatePage for element:text (and any sibling element whose display string is content). If it is deliberately out — because inline maps are the intended route for page prose — the schema comment should say so beside help and subtitle, so the next author finds a decision instead of an absence.

Related: #14253 (three other authored display surfaces with no key; merged as #14381), #14376 (extractor coverage of new key families).

---

Triage — two corrections, and they move the card's centre of gravity

Measured at origin/main b8562ff262.

(a) The face no longer carries submitLabel. It left in @objectstack/spec 17 (#10926, ADR-0049) and the current shape is five keys: title description label placeholder emptyText (translation.zod.ts:927-931). Minor, but the card's premise is a census, so the census should be right.

(b) ⭐ The inline map is not a workaround — it is the ruled route, and this exact question was decided twice.

packages/spec/src/ui/component.zod.ts:1784 declares content: I18nLabelSchema, and the docblock above it (:1773-1783) records why:

I18nLabelSchema rather than a bare z.string() (#5728, named explicitly in the maintainer's ruling because the label-wide widening could not reach it): sys-user.page.ts authors eight element:text nodes whose content is an inline { en, 'zh-CN', 'ja-JP', 'es-ES' } map… The bare string was the declaration disagreeing with the delivered shape.

And the submitLabel retirement one file over (translation.zod.ts:867-876) is the same shape, decided the same way:

the live form surface's submit copy is object-form's submitText (I18nLabelSchema), localizable at its own authoring site, and adding it here would be a face widening.

element:text.content is an I18nLabelSchema localizable at its own authoring site. By the precedent, adding it to the bundle face is a face widening — the thing #10926 declined to do for the identical shape.

⇒ The card's framing ("a missed key… neither named nor excluded") does not survive. What is genuinely missing is the sentence, not the key: content's absence is a decision that was made elsewhere and never written down beside help and subtitle. That is the card's own second option, and the precedent points at it.

Why it is still a decision, and not simply closed

Because the card raises something the two rulings did not weigh. #5728 and #10926 both answered "can this string be localized?" — yes, at its authoring site. This card asks "can it be extracted and counted?" — and the answer is no. Sixteen strings on one page, invisible to os i18n extract, to check:i18n-coverage, and to a translator working from the bundle. That is a new argument against a settled route, measured outside this repo, and it is the maintainer's to weigh rather than the seat's to wave through on precedent.

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

• ① 项目长远合理性(权重 ≥50%,领起推荐) —— 平台已经为「作者写的一段话怎么本地化」定过两次,而且是同一形状:component.zod.ts 的 I18nLabelSchema 键声明为纯字符串,但 3 个平台页在 31 处授权内联多语言 map #5728 把 element:text 的 content 从裸字符串改成 I18nLabelSchema(维护者点名);i18n: component-translation submitLabel copy key lost its only declared carrier when element:form retired (#9249) — decide retire vs re-anchor #10926 把 submitLabel 从 bundle 面退役,原话是「在它自己的编写点可本地化,加到这张面上就是面的加宽」。⇒ ①指向不加键:把同一个字符串交给两条本地化路径,是特例增生而不是收敛。把 content 的缺席写成一条有理由的排除(与 help、subtitle 并列),才是缩小。
• ② 实际业务拉动 —— 有,而且是外部实测的:duly#13 一个页面 16 条字符串,对抽取器、覆盖率门禁和照 bundle 工作的译者全部不可见,已被那个应用记进自己的 untranslatable 台账并按数量钉住。⚠️ 这一条在 component.zod.ts 的 I18nLabelSchema 键声明为纯字符串,但 3 个平台页在 31 处授权内联多语言 map #5728 与 i18n: component-translation submitLabel copy key lost its only declared carrier when element:form retired (#9249) — decide retire vs re-anchor #10926 裁决时都没有被称量过 —— 那两次问的是「能不能本地化」,答案是能;本卡问的是「能不能被抽取和计数」,答案是不能。新论据,不是重提旧案。
• ③ 防 AI 犯错 —— 出错时谁看到什么:一个作者给页面上每一段可见文字找 bundle 键,唯独 element:text 那段找不到,而架构里没有任何东西告诉他这是有意的。help 和 subtitle 都写了排除理由,content 只是不在表里 —— 缺席与遗漏在读者眼里一模一样,他会合理地以为自己漏了什么,或者以为这是个 bug。⭐ 无论方向怎么裁,这一棱都必须被满足:要么补键,要么补那句话。
• ④ 创业阶段不扩散 —— 加一个键是一条永久义务:抽取器、覆盖率门禁、译者工具、向后兼容,外加「同一字符串两条路径谁赢」这个新的冲突面。不新增优于 declare-and-maintain。

推荐:A(卡面的第二个选项)—— 不加键,把排除理由写进 schema 注释,与 help、subtitle 并列,引 #10926 的原话作依据。 ①④同向,③被它满足。
⚠️ 但 A 不解决 ②,必须说清楚:那 16 条字符串仍然对抽取器和覆盖率门禁不可见。所以荐 A 的同时建议另开一张卡问「内联 I18nLabel 映射该不该被 os i18n extract 与 check:i18n-coverage 看见」—— 那才是 ② 真正要的东西,而且它对所有用内联映射的字段都成立,不止 element:text,因此比给一个组件补键更能关掉这一类。⛔ 本席不代开那张卡,那是拿到方向之后的事。
回退:B(卡面的第一个选项) —— 把 content 加进面并在 translatePage 里解析。若维护者认为 bundle 面才是页面散文的正路、内联映射只是过渡形态,走这条。代价:同一字符串两条路径,并且必须一并裁定两者同时存在时谁赢。
置信缺口(本分析看不见什么): 没有量 objectui 侧 pickLocalized 与 bundle overlay 的优先级。若走 B,一个 content 同时带内联映射和 bundle 条目时谁赢,是 B 上线前必须先答的问题 —— 本轮没量,而它决定 B 的真实代价是「加一个键」还是「加一个键 + 一条冲突规则」。

Generated by Claude Code

[os-project-manager]

Maintainer ruling recorded — A: no content key on the bundle face; the exclusion is written beside help and subtitle; the extraction question becomes its own card (#14749)

Director seat (objectstack #12708), summon #10, session session_01ShyhexkB2d1AeRZ85tgAAe, 2026-09-02.

Provenance (who / verbatim / where): maintainer, live PM chat with the director seat, 2026-09-02, replying to decision batch #13 in which this card was item 3 with the recommendation A (fallback B), matching the triage facets on the body. Verbatim reply: 「同意」.

Ruled: A. element:text's content does not join PageTranslation.components. Its localisation route is the inline I18nLabel map at the authoring site, which #5728 (the maintainer named content explicitly) and #10926 (submitLabel retired from the face for the identical shape, "localizable at its own authoring site; adding it here is a face widening") already ruled. The schema comment in packages/spec/src/system/translation.zod.ts gains the sentence that was missing: content is excluded on purpose, beside help and subtitle, citing #10926. Option B (add the key and a precedence rule for the two routes) is not taken.

The card's new argument, that inline maps are invisible to os i18n extract, check:i18n-coverage and translators working from the bundle, is real and unweighed by the earlier rulings; it applies to every inline I18nLabel field, not only this component, so it is carried as its own card: #14749 (domain:cli, pm:queue). duly's 16-string untranslatable ledger is the measurement that card starts from.

Execution: domain:spec lane, S (one comment sentence; no zod change; no changeset beyond what the changeset-presence gate demands). Clause-② no.

State transition, same stroke: needs-user-decision → pm:queue; priority:p2, domain:spec, i18n retained. Ledger: objectstack director seat post #12708, summon #10.

---

Generated by Claude Code