record:details 分区标题存在 title/label 双拼写:三个消费方都读 `title ?? label`,但 spec 只声明 label

[os-zhuang]

实施 #5611(RecordDetailsProps.sections 改对象形式)时实测到的范围外问题,只记录不修。

实测

record:details 分区的标题有两个拼写,三个消费方都做了 title ?? label 或 label ?? title 的容忍读取:

消费方	位置	读法
objectui 渲染器	plugin-detail/src/renderers/record-details.tsx:193	s.title ?? s.label
仓内 i18n 抽取	packages/cli/src/utils/i18n-extract.ts:401	s.label ?? s.title
仓内 lint	packages/lint/src/validate-translatable-sections.ts:329	strName(section.label) ?? strName(section.title)

而授权侧只有一种拼写:

• 仓内全部 5 个真实 section(3 个 showcase 页 + sys-user.page.ts)都写 label,零处写 title;
• objectui 的类型镜像 RecordDetailsComponentProps.sections[] 只声明 label;
• objectui 的 Studio 区块设计器 block-config.ts:312 的 itemFields 也只产出 label。

唯一写 title 的地方是一个合成测试夹具:packages/cli/test/i18n-section-coverage.test.ts:220({ name: 'timeline', title: 'Timeline', fields: [...] })。

#5611 的 PR 据此把 sections[].label 声明为 I18nLabelSchema.optional()、不声明 title —— 按 Prime Directive #12,授权侧的事实拼写是 label,title 属于消费方容忍,不应写进契约固化成第二套正名。

为什么今天不痛、但值得记

今天没有作者写 title,所以没有用户会踩到 —— 故打 finding 不进队列。

值得记的是方向:#5611 之后 title 成为 spec 明确不认的拼写,而三个消费方仍然优先读它(objectui 甚至把 title 排在 ?? 链前面)。这正是 Prime Directive #12 说的"消费方容忍把错误约定化石化成第二套事实契约":一个照着 objectui 源码而不是照着 spec 写元数据的作者,会得到一个 spec 拒绝、渲染器却正常显示的页面。#5068 闸门落地后,这类页面会在 publish 期硬失败,而错误信息指向的是作者从渲染器里学来的拼写。

修复方向

按 #5611 已确立的方向清掉容忍侧,而不是补声明:

1. 仓内两处(i18n-extract.ts:401、validate-translatable-sections.ts:329)去掉 ?? title 分支;
2. packages/cli/test/i18n-section-coverage.test.ts:220 的夹具改写为 label(该夹具用 title 正是为了覆盖这条容忍分支,分支去掉后夹具应一并改拼写);
3. objectui 侧 record-details.tsx:193 去掉 s.title ??(跨仓,需单独 PR)。

注意第 2 条是 #5046 记下的"夹具三分法"里的整体替换类:那个夹具钉的正是要删的那条分支,分支删掉后它会因为"什么都没产生"而继续通过,而不是因为逻辑正确。

[claude]

发现分诊轮判级(#4949 纪律):持有(finding 留)。

为什么还留着:方向上没有疑问 —— 授权侧事实拼写是 label(仓内 5 个真实 section 全写 label,objectui 类型镜像与 Studio 区块设计器也只产出 label),唯一写 title 的是一个合成测试夹具;#5611 的 PR 已按 PD #12 只声明 label、不把消费方容忍写进契约。剩下的是清掉三处 title ?? label 容忍读法,但它跨三个车道:objectui 渲染器(repo:objectui)、packages/cli 的 i18n 抽取(domain:cli)、packages/lint(domain:devx)。为一个零用户影响的清理做 contract-first 拆分(父单 + 三个 sub-issue + Blocked-by:),拆分成本明显大于收益。

重启条件:① 上述任一车道的 PM 下次因别的单动到这三个文件时顺带收(本评论即为那时的依据,三处位置已逐一锚定:plugin-detail/src/renderers/record-details.tsx:193、packages/cli/src/utils/i18n-extract.ts:401、packages/lint/src/validate-translatable-sections.ts:329,外加合成夹具 packages/cli/test/i18n-section-coverage.test.ts:220);或 ② title 拼写首次出现在真实授权面(而非测试夹具)时立即晋级 —— 那意味着容忍读法已经开始生产第二套方言,正是 PD #12 要防的东西。

本评论来自分诊座位 Routine(#5474 试点),不构成认领。

---

Generated by Claude Code

[claude]

Findings triage — re-check of the least-recently-reviewed hold. Verdict: HOLD stands (finding kept). Both restart cells re-verified as NOT triggered.

Reading method matters here. The natural check ("has anyone touched those three files since 08-06?") was attempted with git log --since and discarded: this checkout is shallow, and the log returns the same single boundary commit for every path — the false-reading shape Operational note 6 warns about. The reliable check is presence of the code itself on origin/main, which is what is recorded below.

Cell ① — "a lane collects it while passing through": not triggered. All three tolerance reads are still present verbatim at their anchored positions:

• packages/cli/src/utils/i18n-extract.ts:401 — addSection(index, objectName, s.name, s.label ?? s.title);
• packages/lint/src/validate-translatable-sections.ts:329 — const heading = strName(section.label) ?? strName(section.title);
• packages/cli/test/i18n-section-coverage.test.ts:220 — the synthetic fixture still authors { name: 'timeline', title: 'Timeline', ... }.

Cell ② — "title appears on a real authoring surface": not triggered, and the evidence got stronger. Every record:details sections: array on origin/main was re-read: examples/app-showcase/src/ui/pages/project-detail.page.ts (3), settings.page.ts (2), task-detail.page.ts (3), packages/platform-objects/src/pages/sys-user.page.ts (4). All ~12 section entries write label; zero write title. ⚠️ Note the issue body's figure of "5 real sections" is now stale — the authoring surface roughly doubled since filing and still produced no title. That cuts in favour of holding: the tolerant reads are not, in practice, breeding the second dialect PD #12 warns about — which is precisely what cell ② was written to detect.

One false-positive trap for whoever checks this next: a bare title: grep over *.page.ts hits plenty of lines (page:header titles, related-list titles, sys-user.page.ts:160/:200). Those are different properties. Cell ② is only about title inside a record:details sections[] entry — check the enclosing array, not the keyword.

Hold reason unchanged: the cleanup spans three lanes (repo:objectui renderer, domain:cli extractor, domain:devx lint) for zero user impact today, so a contract-first split costs more than it returns. Restart conditions carry forward verbatim.

本评论来自分诊座位 Routine(#5474 试点),不构成认领。

---

Generated by Claude Code

[os-project-manager]

Findings sweep (maintainer-authorized one-off, 2026-08-07 — registered on #6015): promoted to the queue. title/label dual spelling with three tolerant readers while spec declares label only — the maintainer's standing family default (silently-tolerated second spelling → converge on the declared one); all real authoring already uses label, so the fix is dropping the tolerant reads, zero migration. finding → pm:queue.

---

Generated by Claude Code

[qq9340100]

认领(跨域例外路径 + 跨仓;findings sweep 晋级依据 = 维护者家族缺省「静默容忍的第二拼写 → 收敛到声明拼写」,零迁移 —— 全部真实语料已写 label)

• 会话:session_011M7UwH25Unfi73UHim7ajY(spec 车道执行 PM;dev 为其容器内子代理)
• 分支:本仓 claude/issue-5730-label-spelling;objectui 仓 claude/issue-5730-label-spelling(两仓两分支两 PR,objectui PR 挂 Blocked-by 本仓 PR 语义上不需要 —— 两半独立可合,互不破坏)
• 完整文件面申报(分诊 08-06 评论已逐处锚定,行号以内容为准):
  • objectui packages/plugin-detail/src/renderers/record-details.tsx:193(title ?? label 容忍读 → 只读 label)
  • 本仓 packages/cli/src/utils/i18n-extract.ts:401(s.label ?? s.title → s.label)
  • 本仓 packages/lint/src/validate-translatable-sections.ts:329(同型)
  • 本仓夹具 packages/cli/test/i18n-section-coverage.test.ts:220(合成夹具 title: → label:)
  • ⛔ 不动 packages/spec(RecordDetailsProps 与真实页面的授权形状不符:sections 声明为 string[] 但所有页面授权对象形式,hideFields 完全未声明 #5611 已只声明 label,契约面本就正确)
• 定向在飞检查(cli / devx 车道 + objectui):fix(cli): reserve stdout for the --json payload across the bootSchemaStack family (#6217) #6524(cli,bootSchemaStack stdout)不触 i18n-extract;devx 车道无在飞单触 validate-translatable-sections;objectui 仓无在飞分支触 plugin-detail(ls-remote 核过)。分诊 08-07 复核已确认三处容忍读原样在位、真实语料 ~12 处 section 全写 label 零 title。

---

Generated by Claude Code