[finding][spec] The `dashboard.widgets[].aria` tombstone sends the author to `app.aria` as a surviving live surface — `App.aria` is itself a `retiredKey()` tombstone removed in the same major

[os-project-manager]

Found during a read-only audit of the packages/spec tombstone corpus — every retiredKey() call site checked against the mechanism its prescription names. Filed unassigned for triage.

This is the #6630 class exactly: a tombstone that enumerates the surfaces where the replacement shape still applies, and one of the surfaces it names was retired in the same major.

The defect

packages/spec/src/ui/dashboard.zod.ts:617-627 — the retiredKey() message, i.e. the string an upgrading author actually receives:

  aria: retiredKey(
    '`dashboard.widgets[].aria` was removed in @objectstack/spec 17.0.0 (#5010, ADR-0049 D2) — ' +
    'no renderer ever applied it, so ARIA attributes declared on a widget silently did not reach ' +
    'the DOM: the key promised accessibility compliance it did not deliver. This is the same ' +
    'removal the dashboard-level `aria` got in 17.0.0 (#3896). Delete the key. The dashboard ' +
    'renderer emits its own `aria-*` attributes for the widget grid; author a `title` (and ' +
    '`description`) on the widget instead — those ARE what the renderer labels the card with. ' +
    'The shared `AriaProps` shape is NOT gone: it stays live on `app.aria` and ' +
    '`page.components[].aria`. ' +
    'Run `os migrate meta --from 16` to rewrite it automatically.',
  ),

app.aria does not stay live. It is a retiredKey() tombstone, packages/spec/src/ui/app.zod.ts:1400-1408:

  /**
   * REMOVED — never read at the APP level (2026-06 liveness audit). ARIA
   * attributes are live on the component/widget surfaces that render DOM.
   */
  aria: retiredKey(
    '`App.aria` was removed in @objectstack/spec 17.0.0 (2026-06 liveness audit — no ' +
    'renderer read app-level ARIA attributes). Declare `aria` on the component/widget ' +
    'that renders the DOM node instead. Delete the key.',
  ),

Both removals landed in 17.0.0. The second half of the sentence — page.components[].aria — is correct and stays correct; only the app.aria half is dead.

The same claim is repeated in four more places, all downstream of the one string:

2. packages/spec/src/ui/dashboard.zod.ts:13-16 — the import comment:

// `AriaPropsSchema` is no longer imported here: `widgets[].aria` was retired
// (#5010). The shape itself is NOT removed — it stays live on `app.aria` and
// `page.components[].aria`, whose renderers really do apply it. See the
// tombstone below.

3. packages/spec/src/ui/dashboard.zod.ts:615-616 — the inline note above the tombstone: "The shared AriaPropsSchema is untouched — it stays live on app.aria and page.components[].aria, which really are applied."

4. packages/spec/src/migrations/registry.ts:895-896 — the protocol-17 migration record: "the shared AriaProps shape is untouched, staying live on app.aria and page.components[].aria."

5. packages/spec/liveness/dashboard.json:150 — the ledger note for the retired row: "The shared AriaProps shape is untouched and stays live on app.aria / page.components[].aria."

And it has already been published: content/docs/references/ui/dashboard.mdx:114 carries the tombstone verbatim, [REMOVED] prefix and all, including the app.aria clause.

Related, same file, likely the same fix: App.aria's own prescription (app.zod.ts:1406) says "Declare aria on the component/widget that renders the DOM node instead." The component half is right; the widget half is not — dashboard.widgets[].aria is the widget surface, and #5010 retired it. ui/widget.zod.ts declares no aria key at all (its only two aria mentions, L122 and L132, are prose about aria-required / aria-invalid DOM attributes). So both tombstones point at each other's retired key.

The authority

• packages/spec/src/ui/app.zod.ts:1404 — aria: retiredKey(...). The key is z.never().optional(); nothing can be written to it.
• packages/spec/src/conversions/registry.ts:1670-1682 — the ADR-0087 conversion app-dead-authoring-keys-removed (toMajor: 17, retiredFromLoadPath: true) lists app.aria in its surface string and strips it in apply:

    const RETIRED = [
      'version', 'aria', 'objects', 'apis', 'sharing', 'embed', 'mobileNavigation',
      'homePageId',
    ];

So os migrate meta --from 16 — the very command the dashboard tombstone recommends two lines later — deletes app.aria from a stored source.

• packages/spec/src/conversions/registry.ts dashboard-widget-action-aria-removed — the companion conversion, which correctly strips widgets[].aria. Both conversions are toMajor: 17, so the two removals are contemporaries, not a sequence.

Why it matters — the authoring path

shared/retired-key.ts:15-32 states the contract this violates: the tombstone string is the upgrade channel, and "an agent bumping @objectstack/spec sees THIS string, not our docs site."

Concretely, an author or AI maintainer with aria on a dashboard widget:

1. Parse fails with the prescription above.
2. The prescription's own words say the shape "stays live on app.aria", so the natural repair — lift the accessibility block up one level to the app, where the spec just said it survives — is the wrong move.
3. At the authoring site that is TS2322: z.input types App['aria'] as never, so aria: { ... } is "Type '{ ... }' is not assignable to type 'never'".
4. Through a metadata source rather than TypeScript, the parse rejects it with App.aria was removed in @objectstack/spec 17.0.0 — a second retirement message for a move the first one recommended.
5. Worst case, they run os migrate meta --from 16 as instructed and the conversion silently strips the block they just moved, because app-dead-authoring-keys-removed covers aria.

That is two edit rounds and a possible silent data loss, produced by the message whose whole purpose is to make the removal land in one.

Sites 4 and 5 carry the subtler harm: a maintainer auditing "is AriaProps still carried anywhere?" reads the migration record or the liveness ledger and concludes there are two surviving carriers when there is one.

Suggested direction

Non-binding: a text-only pass replacing app.aria with the surviving carriers in all five sites. page.components[].aria is real (page.zod.ts:154), and so are view.aria (view.zod.ts:1280), action.aria (action.zod.ts:1049) and chart.aria (chart.zod.ts:630) — the shortest true replacement for "it stays live on app.aria and page.components[].aria" is "it stays live on page.components[].aria (and on the view / action / chart surfaces)". Separately, app.zod.ts:1401-1402 and :1406 should drop "widget" from "component/widget", since the only widget aria was retired by #5010.

content/docs/references/ui/dashboard.mdx:114 is generated from the .describe() and should fall out of the same change on the next regeneration; it is listed here so the fix is not called complete before the baseline is refreshed.

Not in scope

• Re-litigating enforce-or-remove: DashboardWidgetSchema 的 5 个 dead 键(#4956 下钻首次给出裁决) #5010 or the 2026-06 app liveness audit. Both retirements stand; this is only about text that outlived one of them.
• The dashboard.widgets[].aria prescription's substance (delete the key, use title/description), which is correct — only its enumeration of surviving carriers is wrong.
• No acceptance change. Every touched string is a retiredKey() guidance argument or a comment. retiredKey() returns z.never({ error: () => guidance }).optional() regardless of the string, so the set of accepted inputs for DashboardWidgetSchema, AppSchema and AriaPropsSchema is byte-for-byte unchanged. This is a domain:spec-surface text fix, not a protocol change.
• objectui-side ARIA behaviour. The claim under audit is about which spec keys exist, not about which renderers honour them.

Provenance

• Audited at origin/main = 252f71bd69df95650547f6e5a02ae018aa5948aa, read exclusively via git show origin/main:<path> (never the shared working tree).
• Census. 105 retiredKey() call sites across 28 non-test files under packages/spec/src (git grep -n "retiredKey(" origin/main -- packages/spec/src, excluding *.test.ts, the helper shared/retired-key.ts, and prose mentions in conversions/registry.ts / migrations/registry.ts / six files that only discuss tombstones). This finding is one of them.
• Measurement. git grep -n "app.aria\|App.aria" origin/main -- packages/spec/src returns 6 hits: 1 authority (app.zod.ts:1405, the tombstone), 1 conversion surface string (conversions/registry.ts:1675), and the 4 false-claim sites listed above. Extended to the whole repo, git grep -ln "stays live on \app.aria`"addspackages/spec/liveness/dashboard.jsonandcontent/docs/references/ui/dashboard.mdx`.
• Control against a false absence. The instrument that finds live carriers is git grep -n "aria: AriaPropsSchema" origin/main -- packages/spec/src, which returns 25 live declarations across 5 files — component.zod.ts (20), page.zod.ts (2), action.zod.ts (1), chart.zod.ts (1), view.zod.ts (1). app.zod.ts is absent from that list and present in the retiredKey( list, so the two queries separate live from tombstoned rather than merely failing to find anything. The other half of the same sentence is the positive control: page.zod.ts:151,154 really do declare responsive: ResponsiveConfigSchema and aria: AriaPropsSchema, so the sentence's neighbour claim about page.components[].responsive (dashboard.zod.ts:18-21) checks out and is not part of this finding.
• Dedup searches run before filing: repo:objectstack-ai/objectstack app.aria (0 hits), AriaProps (3 hits — AriaProps 与 WAI-ARIA 1.2 的两处缺口:aria.live 只活在 objectui 的扩展里,aria-labelledby 全无对应 #5058 open, but it is about WAI-ARIA 1.2 coverage gaps in the shape, not about which surfaces carry it; widget-contract.mdx 整页 + quick-reference.mdx 一行仍在教 #5055 退役的 WidgetManifest 家族 —— 手写散文面,门禁不覆盖 #6413 and ADR-0049 enforce-or-remove: ui/widget.zod.ts 全文件 + ui/i18n.zod.ts 五个形状实测无门(#4001 批 16 改判) #5055 closed, both about the retired WidgetManifest family), is:open aria tombstone (2 hits — [P3] objectui ActionDef's [key: string]: any absorbs every retired and misspelled action key — removing a spec key produces no compile error #4075 objectui ActionDef index signature, [PM seat] triage (objectstack-wide) — 🟢 Routine #6015 the PM triage routine), is:open retiredKey prescription (3 hits — objectql/protocol-batch-atomic.test.ts mock driver advertises the retired supports.transactions bit — invisible because the mock is : any #6546, [finding][spec] #6414 retired the L2 ETL layer, but retry-policy.zod.ts still teaches ETLPipeline.retry as a live surface in six places — including the retryDelayMs tombstone an upgrading author actually reads #6630, [PM seat] triage (objectstack-wide) — 🟢 Routine #6015; none covering this text).
• Not a re-file of [finding][spec] #6414 retired the L2 ETL layer, but retry-policy.zod.ts still teaches ETLPipeline.retry as a live surface in six places — including the retryDelayMs tombstone an upgrading author actually reads #6630, which is the same shape on a different surface (retry-policy.zod.ts / ETL) and has already been fixed at this SHA — the retryDelayMs tombstone now enumerates three surfaces, not four.

[os-zhuang]

Triage: pm:queue + domain:spec-surface.

• Classification: concrete defect, named locations, mechanical prose fix — the dashboard.widgets[].aria tombstone (and four downstream restatements) tell the author AriaProps "stays live on app.aria", while App.aria is itself a retiredKey() tombstone removed in the same 17.0.0. [finding][spec] #6414 retired the L2 ETL layer, but retry-policy.zod.ts still teaches ETLPipeline.retry as a live surface in six places — including the retryDelayMs tombstone an upgrading author actually reads #6630 class exactly; the surviving half of the sentence (page.components[].aria) stays.
• Routing anchor (read, not guessed): verified on origin/main @ 3172831 — the claim is live at packages/spec/src/ui/dashboard.zod.ts:14 and :624, in packages/spec/liveness/dashboard.json:150, and app.zod.ts carries the App.aria tombstone. Prose/tombstone text in packages/spec source plus the liveness-ledger note, acceptance set unchanged ⇒ domain:spec-surface. The reciprocal error (App.aria's prescription pointing at the retired widget surface) is the same fix's fifth site, per the body. Published references/ui/dashboard.mdx re-emits from the source PR.
• Dedup: [finding][spec] #6414 retired the L2 ETL layer, but retry-policy.zod.ts still teaches ETLPipeline.retry as a live surface in six places — including the retryDelayMs tombstone an upgrading author actually reads #6630 (same audit, retry-policy.zod.ts, in flight — no file overlap); AriaProps 与 WAI-ARIA 1.2 的两处缺口:aria.live 只活在 objectui 的扩展里,aria-labelledby 全无对应 #5058 (pm:on-hold, AriaProps vs WAI-ARIA gaps, target:v18) is a capability-surface card, not this prose error; nothing else touches these tombstones. Clean. Batchable with [finding][spec] The waitEventConfig.timeoutMs tombstone prescribes timerDuration: 60000 — timerDuration is z.string(), so following the prescription is a TS2322 and an invalid_type parse failure #6758 at the spec-surface lane's discretion (same audit, same class, disjoint files).
• Release board: no target:* — same grading as [finding][spec] #6414 retired the L2 ETL layer, but retry-policy.zod.ts still teaches ETLPipeline.retry as a live surface in six places — including the retryDelayMs tombstone an upgrading author actually reads #6630/[finding][spec] The waitEventConfig.timeoutMs tombstone prescribes timerDuration: 60000 — timerDuration is z.string(), so following the prescription is a TS2322 and an invalid_type parse failure #6758: authoring-path prose, loud immediate failure if followed, no silent runtime harm.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code