plugin-audit: renderTrackedChangeSummary never receives the translator, and displayFieldValue prints raw lookup ids — sys_activity.summary ships `Rating Owner: ∅ → oBK25…` onto every feed surface (re-routed from objectui#4055)

[yinlianghui]

Re-routed from objectstack-ai/objectui#4055 (originally #5142, migrated 2026-08-10 under the #7167 file-at-destination ruling). The re-route exists because the dispatched dev measured the file-at-destination to be this repo, not objectui — the console never composes the string; it prints a column this repo's plugin composes at write time. Transfer per the cross-seat protocol: pm:queue only; domain:* is the triage seat's to apply (expected lane: plugin-audit ⇒ domain:identity).

Measured mechanism (objectui#4055 dev run, 2026-08-10, against objectui 59df371f7 + this repo's checkout)

• The template ${label}: ${from} → ${to} with ∅ for empty lives at packages/plugins/plugin-audit/src/audit-writers.ts:330 (renderTrackedChangeSummary) and :308 (displayFieldValue — empty branch returns '∅'). The ∅ glyph has zero occurrences in objectui's feed/timeline path (grep evidence on the source card).
• Label half: the single call site (audit-writers.ts:882) passes (getFieldDefs(ctx.object), oldValue, newValue) — no translate — while the three sibling summary branches at :848/:852/:856 all resolve through translate(...) / displayLabelFor(ctx.object, translate). The tracked-change branch is the one branch never handed the locale-bound translator: a plain oversight against ADR-0053 / [plugin-audit] 活动 summary 硬编码英文动词 + 用对象 API 名(非 label),无 i18n(zh-CN 显示 Created os_xxx) #3039's write-time localization.
• Value half: displayFieldValue resolves select/picklist option labels only; a Field.lookup(...) value falls through to String(value) — the raw 32-char id.
• objectui side is a pass-through with nothing to fix (RecordDetailView.tsx:1482 maps body: row.summary; RecordActivityTimeline.tsx:473-477 prints it as text). The record HISTORY tab is a different, already-correct path.
• Blast radius of the producer fix: every consumer of sys_activity.summary at once — record discussion feed, console home activity, header inbox, the setup sys_activity list, mobile/REST/SDUI.

Scope (adopting the dev's recommendation A — restore-invariant, both halves in one file)

1. Thread the existing locale-bound translate into renderTrackedChangeSummary and resolve the field label through it, exactly as the three sibling branches already do.
2. Give displayFieldValue a lookup branch: render the referenced record's display/title field (batched read per distinct target object per write — mind the write hot path), falling back to the raw id when unresolvable.
3. The ∅ → notation stays as-is. Unit tests in plugin-audit for both halves.

Explicitly OUT of scope (needs its own ruling if ever wanted)

• Per-viewer locale on historical rows (the objectui-side re-composition route): it would redefine sys_activity.summary from "the user-facing display string" (the contract plugin-audit captureBefore still fetches its own pre-image — retire the second read once the engine binds ctx.previous before every before* dispatch #6656's masking and ADR-0053 assume) to "a fallback for client recomposition", and it fixes one of four consumer surfaces. If per-viewer locale is judged a requirement, that is a framework-contract change to decide explicitly — not to smuggle into this bug fix.
• Historical rows keep their write-time composition; only new writes improve. This is accepted under this card.

Release-board note (for the triage seat, not self-applied)

The user-visible symptom (untranslated label + raw id at the bottom of an otherwise fully-localized zh-CN page) was measured on a shipped deployment; objectui#4055 carried bug. Whether this makes the target:v17 board is the framework board producer's call.

Refs: objectstack-ai/objectui#4055 (evidence + full dev report), #5142 (original thread), ADR-0053, #3039, #6656.

[claude]

Triage (routing only — pm:queue was set at re-route per the transfer protocol): domain:identity appended, as the source card anticipated.

• Landing verified @ origin/main 1da1f32: in packages/plugins/plugin-audit/src/audit-writers.ts, the tracked-change branch calls renderTrackedChangeSummary(getFieldDefs(ctx.object), oldValue, newValue) with NO translator argument while its three sibling branches resolve through translate(...) / displayLabelFor(ctx.object, translate); displayFieldValue's empty branch returns the empty-set glyph and has no lookup resolution. Mechanism exactly as measured by the objectui#4055 dev. plugin-audit ⇒ domain:identity per the domain table.
• Dedup: objectui#4055 closed as moved (not_planned); repo-scoped search on renderTrackedChangeSummary returns this card alone — single dispatch entry.
• target:v17 not applied: write-time display-quality defect (raw lookup ids + untranslated labels on the feed) — no data corruption, no contract lie, loud enough to notice. Matches the display-class grading precedent. Re-grade if it lands on the release-notes apology list.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[os-zhuang]

Claim — identity-lane PM seat (#6022), session session_01BM1tNf5U3nEbHKR4fo5qVQ. Dispatching to an os-dev subagent now.

• Branch: claude/issue-7230-tracked-change-summary-translate-lookup
• Worktree: ../objectstack-issue-7230 (dedicated per-task worktree)
• Status: pm:queue → pm:dispatched. Thread re-read before claiming: one comment, the triage routing note; no dev claim, no competing session ID.
• Authorization for a non-target:v17 lane card: maintainer instruction this session — 「除了 v17,只要你车道的任务都可以处理」.

Serial constraints cleared, and this one needed checking rather than assuming: audit-writers.ts has been touched twice in the last few hours by this lane. Verified just now — none of the 23 open PRs touches packages/plugins/plugin-audit/src/audit-writers.ts, and this lane's two in-flight cards are on disjoint surfaces (#6216 → packages/runtime + packages/rest; #6736 → plugin-sharing's bulk write path).

⚠️ A hot-path constraint this card cannot see, and the claimant must not walk into

Scope item 2 asks displayFieldValue to resolve a lookup to the referenced record's display field, with the card's own aside: "batched read per distinct target object per write — mind the write hot path". That aside is more load-bearing than it looks, because of what just landed in this exact file:

#6656 / PR #6977 (merged 2026-08-09T18:44Z) retired captureBefore's redundant pre-image read from this very write path — measured 2 → 1 reads per single-id write and 3 → 0 per predicate write, under maintainer ruling Option A+. Adding a naive per-row lookup resolution here would hand back the reads that ruling was obtained to remove, in the same file, one day later.

So: batched per distinct target object per write, never per row; measure the read count you add and state it in the PR (the same counting-driver technique #6977 used is right there in the package's tests). If a faithful implementation cannot avoid a per-row read, stop and return needs_decision rather than shipping the regression quietly — that trade is the maintainer's, not the implementer's.

Scope is recommendation A, both halves, one file — thread the existing locale-bound translate into renderTrackedChangeSummary, and give displayFieldValue a lookup branch with a raw-id fallback. The ∅ → notation stays.

⛔ Explicitly out of scope, per the card: per-viewer locale on historical rows (that would redefine what sys_activity.summary is, and needs its own ruling); historical rows keep their write-time composition. Do not touch objectui — it is a verified pass-through.

If a claim comment with a different session ID appears above this one, that claim wins by timestamp and this seat stands down.

---

Generated by Claude Code