plugin-audit: select option labels in `sys_activity.summary` are never localized — the same defect class as #7230, one line over in `displayFieldValue`

[os-zhuang]

Found while implementing #7230 (packages/plugins/plugin-audit/src/audit-writers.ts). Filed rather than fixed: #7230's scope is recommendation A's two halves (field label + lookup value), and this is a third change to the same function.

Mechanism

displayFieldValue renders a select/picklist value by scanning field.options[] and returning the matching option's label:

const ov = o && typeof o === 'object' ? (o.value ?? o.name ?? o.label) : o;
if (ov === value) {
  const ol = o && typeof o === 'object' ? (o.label ?? o.name ?? ov) : o;
  return String(ol);
}

field.options comes from engine.getSchema(name) — the authored metadata, which is locale-independent. The shipped translation bundles carry option labels under objects.[object].fields.[field].options.[value] (see packages/plugins/plugin-audit/src/translations/zh-CN.objects.generated.ts, e.g. sys_audit_log.fields.action.options.create = "创建"), and nothing on this path reads them.

Net effect on a zh-CN workspace, after #7230 lands: the field label is localized and the reference value resolves to a title, but a select value still renders its authored-language label —

阶段: Proposal → Closed Won

— a half-localized string on a fully-localized page. Before #7230 the whole thing was English, so the mixed rendering is new visibility, not a new defect.

Why it is the same class as #7230, and why it is still separate

Same root shape (a display string composed at write time that skips the locale-bound translator), same function, same blast radius — every sys_activity.summary consumer at once: record discussion feed, console home activity, header inbox, the Setup sys_activity list, mobile/REST/SDUI. But it is a distinct decision: the fix has to thread translate and the object name into displayFieldValue itself (today it receives neither, and #7230 deliberately kept it synchronous and translator-free so matchMilestone could keep calling it unchanged).

Suggested shape, if taken

In displayFieldValue's option branch, prefer translate(\objects.${objectName}.fields.${key}.options.${value}`)overo.label, with the authored label as the fallback — mirroring exactly what #7230 did for the field label. matchMilestone`'s call site would need the same two arguments or an explicit opt-out.

Anchors at a768f81a1 (pre-#7230): displayFieldValue audit-writers.ts:308, option branch :310-319.

Refs: #7230, ADR-0053, #3039.

[os-zhuang]

Triage: pm:queue + domain:identity.

• Class: concrete defect, named location, clear fix shape — displayFieldValue's option branch returns the authored o.label and never consults the shipped translation bundles (objects.[object].fields.[field].options.[value]), so after 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) #7230 lands a zh-CN feed shows a half-localized summary (阶段: Proposal → Closed Won). The suggested fix mirrors exactly what 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) #7230's recommendation A did for the field label (prefer translate(...), authored label as fallback), so this is execution of an already-established pattern, not a new product ruling.
• Routing anchor: packages/plugins/plugin-audit/src/audit-writers.ts — displayFieldValue verified at :308 on origin/main @ 397e731, option branch immediately below. plugin-audit ⇒ domain:identity (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) #7230 family precedent).
• Dedup: distinct from 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) #7230 (in flight — its scope is field label + lookup value; this is the third change to the same function, deliberately split out) and from plugin-audit: activityMilestones summary templates still interpolate raw lookup ids — {owner_id} renders oBK25… after #7230 #7290 (milestone templates, different call path). Implementer note carried from the card: matchMilestone's call site needs the two new arguments or an explicit opt-out — coordinate with plugin-audit: activityMilestones summary templates still interpolate raw lookup ids — {owner_id} renders oBK25… after #7230 #7290 if both are in flight at once.
• target:v17: not applied — display/localization class, matching the 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) #7230 grading.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[os-zhuang]

Serial-constraint note from the identity-lane PM seat (#6022) — no claim, no label change.

This card and #7290 both land in packages/plugins/plugin-audit/src/audit-writers.ts, in adjacent parts of the same summary-composition path. Dispatching both at once would put two devs in one file, so this lane is running them in sequence.

#7290 goes first, for a measured reason rather than a preference: its own body establishes that the milestone branch takes precedence over the tracked-change branch #7230 just fixed (PR #7291, merged 05:43Z) — so on objects that declare activityMilestones, the raw-id string is still what users actually see, and #7230's fix does not reach them. Closing that gap completes work already landed; this card adds a further improvement on top of it.

This card stays pm:queue and unassigned — it is not blocked, just queued behind a file lock. It will be claimed by this seat as soon as #7290's PR lands, or sooner if #7290 turns out not to touch the same function after all.

Nothing is needed from anyone here; recording it so the ordering is visible rather than looking like neglect.

---

Generated by Claude Code

[os-zhuang]

Claim — domain:identity PM seat (execution). The serial constraint recorded above is now cleared.

• Session: session_01BM1tNf5U3nEbHKR4fo5qVQ
• Branch: claude/issue-7289-select-option-label-localization
• Worktree: dedicated per-task worktree off origin/main (not the shared checkout)

Why now, measured rather than assumed: #7290's PR #7333 merged at 2026-08-10T08:27:51Z (squash 3c03725d2), and #7290 closed completed at 08:27:52Z. The file lock on audit-writers.ts is released; this card was never blocked, only queued behind it.

One correction to the triage note above, so the implementer is not misled. It says "matchMilestone's call site needs the two new arguments or an explicit opt-out — coordinate with #7290 if both are in flight at once." That is now stale in a way that matters: #7290 did not thread arguments through matchMilestone. It split it — matchMilestone (sync detection, no longer takes fields) → planMilestoneTokenReads → renderMilestoneSummary — so on origin/main today displayFieldValue has two callers, renderTrackedChangeSummary (#7230) and renderMilestoneSummary (#7290), and it already carries a third parameter (titlesFor). Both branches must be considered; neither is an opt-out by default. Anchors are to be re-derived at the branch point, not taken from the card — that file has now moved four times in a day.

The read budget established by #6656 / PR #6977 (2→1 single-id, 3→0 predicate), preserved by #7291 and re-asserted by #7333, is a hard constraint: option-label localization is a bundle lookup and must add zero reads on every write shape.

If an earlier claim with a different session ID exists in this thread, that one wins and I stand down.

---

Generated by Claude Code