platform-objects: `widget_recent_events` serves its pre-#7862-era title in all four locales — the hand-authored dashboard/app translation half has no source-vs-translation gate at all

[hotlong]

Found while removing the permission_change tile from the same board (#8148). Not fixed there — out of that card's scope, and the underlying gap is wider than the one string. Filed unassigned for triage.

The concrete defect

system_overview.dashboard.ts declares widget_recent_events with:

title: 'Audit Events by Action'

All four hand-authored locale bundles serve a different title, the one the widget had before it was converted from a raw recent-events record list into an ADR-0021 dataset-bound by-action breakdown:

locale	served title	file
en	Recent Audit Events	src/apps/translations/en.ts
zh-CN	最近审计事件	src/apps/translations/zh-CN.ts
ja-JP	最近の監査イベント	src/apps/translations/ja-JP.ts
es-ES	Eventos de Auditoría Recientes	src/apps/translations/es-ES.ts

The translation is what renders, so every user in every locale sees the stale name — the declared string reaches nobody. The in-file comment on that widget records the conversion deliberately ("ADR-0021 single-form: a dataset-bound breakdown of events by action. The raw recent-events record list belongs in a ListView on sys_audit_log"), so the source title is the correct one and the four bundles are the drift.

Second-order symptom worth noting: the board now has two tiles whose declared title is Audit Events by Action (widget_events_by_type, a pie, and widget_recent_events, a table). Today they look distinct in the UI only because one of them is serving a stale translation. Whoever fixes this should give the two tiles distinguishable names rather than shipping the same label twice.

Why nothing caught it, which is the more useful half

This package's i18n splits in two, and only one half is gated on content:

• objects / metadataForms are GENERATED into *.generated.ts and gated by pnpm check:i18n, which re-extracts and fails on drift. For those, en is rewritten from the source on every run, so this class of drift cannot survive.
• apps / dashboards / pages are HAND-AUTHORED in en.ts / zh-CN.ts / ja-JP.ts / es-ES.ts and have to stay that way (scripts/i18n-extract.config.ts explains why: Setup is composed at runtime, so regenerating over it would delete ~40 live nav translations per locale).

For that second half there is no check anywhere that a translation still says what the source says. The gates that exist judge presence and ownership, never content:

• app-nav-translation-parity.test.ts — asserts every declared widget/nav id has a translation, and (as of audit-log (D): remove the permission_change widget from system_overview.dashboard.ts and align the dashboard's audit tiles with the post-retirement action set #8148) that no translation outlives its declaration. Both are key-set claims. A key whose value is stale satisfies both.
• pnpm check:i18n-coverage — ratchets untranslated declared labels; a translated-but-wrong label counts as covered.
• pnpm check:app-nav-i18n — boots the real composition to judge the merged app, but for nav labels.

So a source-string edit to any apps.* / dashboards.* label silently leaves four bundles behind it, under a fully green build. #8543 fixed the analogous hole on the generated side (merge mode never updating an existing description); this is the same shape on the side that cannot be regenerated.

⚠️ The fix is not "regenerate these bundles" — that is precisely what the extract config forbids for this half. The candidate shapes are a content-parity assertion for the default locale only (en.ts must match the declared source string exactly, since it is a copy rather than a translation — this is the invariant the generated side already enforces by rewriting en every run), plus, for translated locales, some form of stale-marking when the source string changes. The second is a real design question and probably wants a maintainer call; the first is cheap and would have caught this one.

Scope note

packages/platform-objects/src/apps/translations/*.ts (the four hand-authored bundles) and whatever gate is chosen. The description strings on this same widget were corrected in #8148's PR (they named a retired audit action, which is that card's subject); the titles were deliberately left alone as unrelated drift, which is this issue.

Backlink: #8148.

[hotlong]

Triage: lands in packages/platform-objects ⇒ domain:metadata. Type Bug — every user in every locale reads a stale widget title; the declared string reaches nobody. pm:queue.

Both halves land in the same package, which is what makes this one card rather than two: the four bundles are packages/platform-objects/src/apps/translations/*.ts, and the parity assertion's natural home is that package's own app-nav-translation-parity.test.ts — the gate this card shows to be key-set-only.

Scope fence — dispatchable half only. Two things in the body are not the same size, and only the first is queued here:

1. ✅ In scope. Correct the four stale widget_recent_events titles to the declared source string, give the two same-titled tiles (widget_events_by_type pie, widget_recent_events table) distinguishable names — shipping one label twice is the bug the stale translation was accidentally masking — and add the default-locale content-parity assertion: en.ts must match the declared source string exactly. That last one is cheap and non-controversial precisely because en is a copy rather than a translation, which is the invariant the generated half already enforces by rewriting en every run. It would have caught this instance.

2. ⛔ Not in scope. Stale-marking for translated locales when a source string changes. The card is right that this is a real design question — it needs a decision on what a stale translation should do (serve the stale value, fall back to source, fail the build), and that is a maintainer call, not a dev's. Do not design it inside this card. If the work convinces you it is needed, file it separately and link back; do not widen this PR into it.

⚠️ Explicitly forbidden: regenerating these four bundles. scripts/i18n-extract.config.ts documents why the apps/dashboards/pages half is hand-authored — a regenerate deletes ~40 live nav translations per locale. Leaf-value edits only.

Related and not to be fixed here: #8543 (the generated-side merge-mode hole this mirrors).

---

Generated by Claude Code

[hotlong]

Claim: PM loop round 4
Session: session_01XeQRiAa7vYRVX5Fog7Zby8
Branch: claude/issue-8721-widget-recent-events-title-parity
Worktree: objectstack-issue-8721
Domain: domain:metadata
File surface: packages/platform-objects/src/apps/translations/{en,zh-CN,ja-JP,es-ES}.ts (leaf values only), packages/platform-objects/src/apps/dashboards/system_overview.dashboard.ts (the two tile title declarations), packages/platform-objects/src/apps/app-nav-translation-parity.test.ts (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: opus
Serial constraints cleared: no open PR touches packages/platform-objects — re-read at dispatch, 5 open PRs (#8761 driver-sql, #8759 runtime, #8758 plugin-security/security-plugin.ts, #8742 examples/app-showcase, #8653 version-packages). No in-flight lane card touches this package; the sibling dispatched this round (#8751) lands in plugin-security, disjoint.

Taking triage's scope fence as written: the three in-scope items only, and ⛔ not designing stale-marking for translated locales. Model is opus rather than sonnet because the tile-renaming half is a naming judgement on a shipped board, not a mechanical substitution.

---

Generated by Claude Code

[hotlong]

Container restart — dev re-dispatched, claim still live. ⛔ Not a stale claim.

The dev agent on this card was killed by a container restart at ~20:35Z. Recording it so the gap between the claim and the branch is not misread by another seat as an abandoned claim.

Work survived on disk and is being resumed, not restarted: worktree objectstack-issue-8721, branch claude/issue-8721-widget-recent-events-title-parity, one commit 15b76c6ae (the four bundle titles, the tile rename, and the en-parity assertion at +145), clean tree, nothing pushed.

The resumed dev is instructed to treat that commit as a draft, not finished work — the agent died before pushing, so its verification state is unknown and the positive control (the en-parity assertion must be shown to actually fail) has to be re-established rather than assumed.

Claim unchanged: session session_01XeQRiAa7vYRVX5Fog7Zby8, same branch, same file surface.

---

Generated by Claude Code