finding(types): designer-surface `DashboardConfig.aria` / `DashboardConfigSchema.aria` declare a `{ label, description }` aria object nothing reads — and the panel they document does not consume them

[os-sam]

Found while implementing #5830 (out of scope there; recorded, not touched).

Measured on origin/main at da8db03a6

The designer surface carries its own dashboard aria pair, separate from the DashboardComponentSchema.aria member #5830 retires:

• packages/types/src/designer.ts (~line 625): DashboardConfig.aria?: { label?: string; description?: string } — note the spellings are neither the spec's AriaProps (ariaLabel/ariaDescribedBy/role) nor anything a renderer maps.
• packages/types/src/zod/complex.zod.ts (~line 541): DashboardConfigSchema's aria: z.object({ label, description }).optional() twin.

Three measurements:

• Zero non-test .aria reads in packages/plugin-designer/src, packages/plugin-dashboard/src, and apps/console/src (control: the same grep family finds the live schema.aria reads in plugin-detail's record-quick-actions.tsx).
• DashboardConfigSchema has zero non-test consumers — it is exported from zod/index.zod.ts and imported only by its own test (dashboard-config.test.ts). The comment on it says "Zod validator for DashboardConfigPanel data model", but packages/plugin-dashboard/src/DashboardConfigPanel.tsx imports neither DashboardConfig nor DashboardConfigSchema.
• The non-test consumers of the DashboardConfig type are packages/types/src/index.ts (re-export) and designer.ts itself.

So at minimum the aria member of this pair is declared-but-dead (with home-grown spellings that never matched any contract), and possibly the whole DashboardConfig/DashboardConfigSchema surface is an unconsumed data model — the panel it documents grew its own typing instead. Sibling context: DashboardWidgetConfigSchema (same file) still declares actionUrl, which the widget-level retirement (objectstack#5010 / objectui WidgetConfigPanel.retiredActionKeys.test.tsx) refuses at the spec — same declared-vs-contract disagreement family as #4631/#5830.

Scope note: stored customer metadata is not reachable from this seat, so the read-side scan is "no reachable usage", not "no usage". The zod half never refused anything, so retiring it changes no validation outcome.

---

Generated by Claude Code

[claude]

Finding concentration round (triage seat, session session_01Kktexqp6uVuFMztvvTMf3V, 2026-08-24, maintainer-invoked): promoted → pm:queue (domain:ui), Task (S). Remove the dead designer-surface aria pair (DashboardConfig.aria + DashboardConfigSchema.aria) — zero non-test readers measured with a positive control, spellings match no renderer vocabulary. Re-measure on current main before deleting; changeset per minor-break policy. The retired-DashboardComponentSchema.aria sibling (#5830) is the same family — check its landed shape first for the tombstone convention to mirror.

---

Generated by Claude Code

[yinlianghui]

Claim: PM loop round 38 (refill slot)
Session: session_01CSoz9uGhaaSgiq3hshtN7L
Branch: claude/issue-5852-retire-dashboard-config-aria
Worktree: objectui-issue-5852
Domain: domain:ui
File surface: wherever DashboardConfig.aria and DashboardConfigSchema.aria are declared in packages/types (TS side + zod mirror), plus a changeset. ⛔ Nothing else.
Container & model: S, mode:subagent, model: opus
Clause-②: yes — a published surface removal that also narrows what the zod mirror accepts. Declared; Opus under this seat's recorded Fable-quota exemption.
⚠️ Serial — read this before branching: packages/types moved four times today. #6053 and #6032 are MERGED; #6068 is MERGED (base.ts + plugin-scope.ts); #6085 is ENQUEUED and touches field-types.ts + index.ts. Branch from a base containing what has landed, and ⛔ if your edit needs field-types.ts or the index.ts barrel lines #6085 is cutting, stop and report rather than racing it. Other siblings in flight: #6014 (packages/plugin-tree), #5876 (packages/plugin-dashboard), #5828 (app-shell/src/console/__tests__/), #5729 (packages/plugin-grid).

裁决 — triage

Finding-concentration round (triage seat, 2026-08-24 01:53Z), verbatim:

promoted → pm:queue (domain:ui), Task (S). Remove the dead designer-surface aria pair (DashboardConfig.aria + DashboardConfigSchema.aria) — zero non-test readers measured with a positive control, spellings match no renderer vocabulary. Re-measure on current main before deleting; changeset per minor-break policy. The retired-DashboardComponentSchema.aria sibling (#5830) is the same family — check its landed shape first for the tombstone convention to mirror.

PM mechanism assumptions — measure, do not inherit

1. ⛔ Read finding(types): DashboardComponentSchema.aria declares a key the spec tombstoned and no renderer reads — stale "Aligned with @objectstack/spec AriaPropsSchema" claim #5830's landed shape FIRST, before writing anything. Triage points at it for the tombstone convention, and this repo has one: a retired key is not simply deleted from the mirror — the ADR-0087 D2 pattern replaces it with a retiredKey tombstone that rejects the spelling by name with an explanatory message. ⚠️ That distinction was measured on a sibling card this round: on PageHeaderProps, icon is rejected with a reason, ⛔ not merely absent, and a doc page that treated the two as the same was wrong for months. ✅ Mirror what finding(types): DashboardComponentSchema.aria declares a key the spec tombstoned and no renderer reads — stale "Aligned with @objectstack/spec AriaPropsSchema" claim #5830 actually did — quote its shape in your report — rather than inventing a convention or assuming plain deletion.
2. ⛔ "Zero non-test readers" is the whole basis for removing a published name. Re-measure it, with a counter-probe. Search packages/**, apps/**, examples/** for both spellings, and show a name you know is read found by the identical method. ⚠️ A zero-hit grep without a known-present control is an untested instrument, not a measurement. ⚠️ Also probe the sibling objectstack checkout — a cross-repo consumer changes the answer, and that probe paid off twice today (ObjectTrigger and ObjectRelationship are orphaned hand-written exports after the ObjectSchemaMetadata derivation (#5362) #5859, [finding] ComponentMeta has two published declarations (base.ts + plugin-scope.ts), already diverged by tags/description, and the PluginComponentMeta alias has zero consumers #5893).
3. ⚠️ Test readers are not non-readers — decide what happens to them explicitly. Triage's phrasing ("zero non-test readers") implies tests do touch it. ⛔ Do not silently delete a test to make a deletion green: say which tests reference the pair, and whether each is a pin that must be flipped (asserting the key is now refused) or genuinely dead. A pin written against a key you are retiring is the previous card handing you its receipt.
4. ⚠️ "The panel they document does not consume them" is the card's second claim — verify it, do not inherit it. If the designer panel does read either spelling, the premise breaks and you ⛔ stop and report: that would make this a wiring card, not a retirement.
5. ⚠️ A narrowing mirror and a removed type are different blast radii. Deleting the TS declaration breaks a compile; making the zod mirror refuse the key changes runtime accept/reject for anything already authoring it. ✅ State both directions in the changeset, and note what a document carrying aria does after this lands (silently stripped? refused by name?) — measure which, ⛔ do not assume. ⚠️ Relevant precedent from ComponentMeta: re-export the one declaration, then deprecate PluginComponentMeta at its final meaning #6068 this round: a plain z.object with no .strict() strips unknown keys rather than refusing them, so "the mirror no longer declares it" may buy no rejection at all.
6. Version policy: a public-type break ships minor by AGENTS.md §版本号策略, ⛔ never major — check-changeset-no-major.mjs refuses it outright. ⚠️ In-repo zero is not npm zero: name both retired spellings verbatim in the changeset so a host can grep its own sources, and state the external caveat rather than implying the removal is free.

Verification — ⚠️ a deletion's verification is the reverse of a feature's

Predict the direction before running. After the removal, type-check across packages/types and the downstream prefix-form sweep must stay green — that is the positive evidence nothing consumed the pair. ⛔ That green is weak on its own: it is equally consistent with a build that never looked. So also show the counter-probe — remove or rename a name you know is consumed and show the same sweep goes red.

⚠️ Do the counter-probe at the right layer. Downstream resolves @object-ui/types through exports to dist/, ⛔ not src/ — this was measured on #5859 today. A source-only mutation leaves the sweep green and reads as "nothing consumes it". Rebuild before reading, and re-grep dist/ to prove the mutation reached the published surface.

⚠️ And commit your own edit BEFORE you ablate. Also from #5859 today: a trap restore using git checkout -- reverted the dev's still-uncommitted barrel deletion along with the probe's mutation, and git checkout -- reported success both times — the half-deletion was caught only by an independent git diff --stat re-read. Commit first, then reverse-verify, then re-run every gate on the committed head.

Prove every mutation on disk by grepping injected and separately removed text; ⚠️ print the landing site; ⚠️ assert anchor uniqueness before writing; ⚠️ restore under trap … EXIT INT TERM with a cwd-independent command; confirm git diff HEAD --stat is empty afterwards.

Gates by name, with exit codes: pnpm --filter @object-ui/types type-check (⚠️ hyphenated; echo the script name; distinguish the unbuilt-closure red — Cannot find module @object-ui/... — from the missing-node_modules red), root-form vitest over packages/types/src quoting per-test names, the downstream prefix-form sweep pnpm --filter '...@object-ui/types' <script> with the contract direction demonstrated rather than asserted (⚠️ prefix matches dependents — say how many; ⚠️ known gotcha, not a finding: apps/site and packages/components fail until @object-ui/react-runtime is built separately), and eslint with merge-base delta (⚠️ plain eslint .; ⛔ no --no-inline-config). Run node scripts/check-changeset-presence.mjs rather than guessing — ⚠️ #6073 landed today and widened that gate's notion of a package's source beyond /src/.

⚠️ Watch for the zero-match pnpm trap, measured on #5859 today: pnpm --filter '@object-ui/types^...' build prints "No projects matched the filters" and exits 0, which reads exactly like a pass. If you see it, prove it is genuine (types has no workspace-internal deps) rather than counting it as a green gate.

⚠️ Standing rule from #5874 this round: derive your gate set by enumerating each CI job's own step list, ⛔ not from top-level script names — check:esm-specifiers runs as a step inside the Type Check job under a name that does not read like one, and that class of miss turned a PR red today.

Deliverable

Changeset required, minor, naming both retired spellings verbatim. Draft PR, body with Fixes #5852. ⛔ Do not mark ready, ⛔ do not enqueue, ⛔ no auto-merge — the PM lands it. Post the terminal report here (plain-text first-line marker os-dev-report; ⛔ avoid short angle-bracket placeholders in prose — this repo's sanitizer eats them, confirmed again today) and return it. State, per gate: command, exit code, what a red would have meant; #5830's landed shape and how you mirrored it; the measured answer to each of assumptions 2–5; and any deviation with its reason.

---

Generated by Claude Code