spec: retire `ChartConfigSchema.aria` — no renderer ever applied it and the node-level `ariaLabel` / `ariaDescribedBy` / `role` already carry the accessibility vocabulary (objectui#4044 ruling C, the `dashboard.aria` precedent)

[os-tesla]

Filed by the director seat on the maintainer's ruling for objectui#4044 (decision batch #118 item 2, 2026-09-12: 「其他同意」 to the seat's recommendation C — judge the protocol wrong for this one key). ⛔ packages/spec is the domain:spec seat's; the director seat writes no repo file.

Governing text

• Maintainer standing ruling (2026-09-10, verbatim): 「我们的项目以objectstack 协议为准，文档应该以实际实现为准。协议不正确的应该先修改协议。」
• objectui#4044 ruling (5617466747) — implementation follows the protocol for DashboardWidget.chartConfig; PR objectui#9202 (merged 2026-09-11) forwarded 9 of the 10 DO-NOW keys. The one it could not forward is aria, and the seat stopped and reported (objectui#4044 comment 5639000150).
• Precedent, verbatim from objectui/packages/types/src/complex.ts:1915-1925: "the spec removed dashboard.aria at the POST /data/sharing/rules 绕过 SharingRuleSchema:criteria 缺失或拼错静默变成"共享该对象全部记录",与 ADR-0049 "never seeded as a permissive match-all" 直接冲突 #3896 audit close-out (no dashboard renderer ever applied it), so DashboardSchema.shape.aria is a tombstone that refuses any value" — pinned by __tests__/dashboard-aria-retired-contract-twins.test.ts.

The defect in the protocol

ChartConfigSchema.aria (packages/spec/src/ui/… — locate by symbol) declares a nested accessibility object on a chart's configuration. Measured on objectui origin/main (5639000150): no renderer reads it — ChartRenderer drops every prop but schema / onChartClick, AdvancedChartImpl declares no aria prop, normalizeChartSchema names it nowhere; forwarding it nested or flattened changed nothing in the DOM (lit control: description sets aria-label / role='img' on the same node). Meanwhile the platform already carries ONE accessibility vocabulary at the node level — BaseSchema.ariaLabel / ariaDescribedBy / role, converted to DOM attributes by SchemaRenderer.resolveAriaProps everywhere. chartConfig.aria is a second, block-local spelling of the same thing, with no reader, and giving it a reader would create two accessible-name sources on one element that then need a precedence rule.

⇒ Long-term shape: one node, one accessibility vocabulary, at the base node. The key is retired the way dashboard.aria was.

What to do

1. Retire ChartConfigSchema.aria: tombstone that refuses any value with a message naming the flat trio on the node (ariaLabel / ariaDescribedBy / role), following the dashboard.aria shape.
2. ADR-0087 semantic migration entry: chartConfig.aria.label → node ariaLabel, chartConfig.aria.describedBy → node ariaDescribedBy (adjust to the key's actual members), lossless where the members map one-to-one; a structured TODO for any member that does not.
3. Liveness ledger row for widgets.chartConfig updated to the measured truth (this key never live); #17385 (xAxis / yAxis / series vs dataset derivation) is the sibling protocol card on the same row — coordinate, ⛔ do not fold.
4. major changeset (a declared key stops parsing), Clause-②: no (narrowing).
5. objectui#4044 is pm:blocked on this card; when it lands, objectui's contract twins gain the tombstone pin and the dashboard aria question closes.

Refs

objectui#4044 · objectui#9202 · objectui#5830 / #3896 (the dashboard.aria retirement) · #17385 · packages/types/src/complex.ts:1915-1925 (objectui)

Filed by the director seat, session_01QsCVSivtpwT6ZXs5Rtvqxe, 2026-09-12T02:1xZ, with Claude Code.

---

Generated by Claude Code

[os-warren]

Claim: domain:spec execution PM seat — session_01KB5PFtxuy1x3dcR5gxudx6, claimed 2026-09-15T13:47Z.
Branch: claude/issue-17751-chart-config-aria-retired
Clause-②: yes
Thread-read: this card carries zero comments; this claim opens the thread.

Labels written first (pm:dispatched on, pm:queue off, assignee set) and read back at 13:47:53Z; this comment is the second half of the atomic pair.

Clause-② = yes. Retiring a declared key on a published packages/spec schema narrows the accept set of a shipped contract, so the delivering PR and this card both carry needs:contract-review for the duration, and the gate clears from both carriers in one stroke only against an at-tier PASS naming the landed head.

Premise re-check, run before the claim, on origin/main @ 0865f41443 (2026-09-15T13:4xZ)

⛔ Not read from the card's text — every row measured:

the card asserts	measured
ChartConfigSchema.aria is still declared	packages/spec/src/ui/chart.zod.ts:630 — aria: AriaPropsSchema.optional().describe('ARIA accessibility attributes') ✅
(control, same file, same instrument)	showLegend → 2 hits ⇒ the file reads; the hit above is not a broken pattern
no in-flight writer on that file	last touch e6965dd7ee (#18039) ✅
objectui#4044 is the consuming card	open ✅
objectui#9202 forwarded 9 of 10 DO-NOW keys	merged ✅
the dashboard.aria / app.aria precedent exists	app.zod.ts:1571 — aria: retiredKey(…) ✅

⭐ Three things the card does not spell out, found by the premise check — they ride the dispatch order

1. The precedent is not just prose, it is a live tombstone plus a pin. packages/spec/src/ui/app.zod.ts:1571 already retires app.aria through retiredKey(), and packages/spec/src/ui/aria-carrier-tombstones.test.ts exists specifically to pin that class. ⇒ this retirement joins an existing pin family; it does not invent one. (Control: retiredKey( appears 91 times under packages/spec/src/ui/, so the mechanism is well-worn.)

2. aria has two alias spellings that must be handled with it. chart.zod.ts:557 maps:

   accessibility: 'aria', ariaProps: 'aria',
   

   ⇒ retiring aria while leaving those two alias entries pointing at it would route an author who typed accessibility onto a retired key. The round must decide and state what happens to both. ⛔ Not a detail to discover mid-implementation.

3. The node-level vocabulary that replaces it is the prescription's target. BaseSchema.ariaLabel / ariaDescribedBy / role, converted by SchemaRenderer.resolveAriaProps. The tombstone's guidance has to name that, or an author loses the migration path.

Ruling carried into the order, verbatim and untranslated

Maintainer, decision batch #118 item 2, 2026-09-12 — 「其他同意」 to the director seat's recommendation C (judge the protocol wrong for this one key). Standing principle cited by the card, also verbatim:

我们的项目以objectstack 协议为准，文档应该以实际实现为准。协议不正确的应该先修改协议。

⇒ This card is ruled execution, ⛔ not a proposal. The decision that chartConfig.aria should go has been made; the round implements it and does not reopen it.

Concurrency

Nothing else is in flight in this lane. packages/spec/src/ui/chart.zod.ts is not on the seat's hot-file serial queue (that queue holds stack.zod.ts, held by #18202). os-verify-lock.sh --status read lock free, queue empty at 13:4xZ ⇒ arrival depth 1, under LOCK_DEPTH_HOLD 2.

---

Generated by Claude Code

[os-warren]

os-dev-report

{
  "issue": 17751,
  "status": "done",
  "branch": "claude/issue-17751-chart-config-aria-retired",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18300 (draft, head 16a8de837aec5791ba8976e80272f066bd818118)",
  "session": "session_01KB5PFtxuy1x3dcR5gxudx6",
  "premise_still_valid": true,
  "summary": "Retired ChartConfigSchema.aria as a retiredKey() tombstone, joining the existing aria-carrier family (app.aria, dashboard.aria, dashboard.widgets[].aria) rather than inventing a route. Registered as TWO retired keys, ui/ChartConfig:aria and ui/ReportChart:aria, because ReportChartSchema is a ChartConfigSchema.extend(...) and an extension copies the retired property into its own walked shape; both under major 18 per the launch-window convention its neighbours state. Added the D2 conversion chart-config-aria-removed (toMajor 18, retiredFromLoadPath) stripping all three authored coordinates (dashboards[].widgets[].chartConfig, reports[].chart, reports[].blocks[].chart; fixture expects 3 notices), wired its id into MIGRATIONS_BY_MAJOR[18].conversionIds and extended that step's rationale. The ledger row stays dead with a REMOVED note and verifiedAt 2026-09-15; its undrilled-containers.baseline.json row is deleted, which check:liveness itself demanded. Changeset is minor with the ADR-0087 disposition marker. PREMISE RE-MEASUREMENT, every row taken first-hand in the worktree, not read off the card: (1) chart.zod.ts:630 still declared 'aria: AriaPropsSchema.optional()' (control showLegend = 2 hits in the same file); (2) at objectui pin 53ded82bf7a4, chart-presentation.ts mentions aria only in prose, its own docblock calling it 'the one declared key with no reader at all' (control showLegend = 3 hits, same file); (3) at objectui HEAD ff1d5ea8d every .aria hit on the chart paths is a NEGATIVE pin, against a control read that does exist one surface over (plugin-view/src/ObjectView.tsx, 1 hit, the VIEW's); (4) precedent confirmed live at app.zod.ts:1571, dashboard.zod.ts:829 and :1190, with the pin family already at src/ui/aria-carrier-tombstones.test.ts (control: 'retiredKey(' = 91 hits under packages/spec/src/ui/). Assignee was already set to os-warren by the dispatch and I did not touch it; the newest Claim: comment (5681276457) names my branch. I hung NO labels: the PR carries none, per the seat's mid-run correction.",
  "alias_decision": "BOTH alias entries (accessibility to aria, ariaProps to aria) moved out of the aliases table and into guidance, carrying the retirement text instead of a rename. This turned out NOT to be an ambiguous judgement, so it is not a stop-and-report: the alias map's own rules decide it, and I ablated the decision in both directions. (a) LEAVING THEM IS MECHANICALLY REFUSED. Restoring the two entries turns src/shared/alias-integrity.test.ts red, naming both by name: 'ui/chart.zod.ts:540 - \"this chart config\": `accessibility` to `aria` - `aria` is a tombstone; it accepts nothing' and the same line for ariaProps. The alias table is consulted BEFORE the edit-distance fallback and bypasses the acceptsNothing filter that keeps tombstones out of the candidate list, so an alias left in place routes an author onto the one key the shape is now guaranteed to reject. (b) DELETING THEM WITH NO REPLACEMENT LOSES THE MIGRATION PATH. In the same ablation run the author who writes 'accessibility' gets a bare 'Unrecognized key(s) on this chart config...' with no retirement text in it, because knownKeys excludes tombstones from the distance candidates and nothing else points anywhere. guidance is the documented channel for exactly this (StrictObjectOptions: 'tombstones for retired keys ... an entry here suppresses the rename suggestion'), and alias-integrity's companion rule 'no guidance key is itself a declared key' is satisfied: accessibility and ariaProps are not declared, while aria (which stays in the shape as a never) is deliberately NOT given a guidance entry.",
  "shipped_prose_before_after": "THE CARD'S SHORTHAND WAS FALSE FOR A packages/spec TOMBSTONE, and this is the one correction that changed what shipped. The card and the dispatch order both describe the replacement vocabulary as BaseSchema.ariaLabel / ariaDescribedBy / role, converted by SchemaRenderer.resolveAriaProps. Verified on the tree before writing: there is NO BaseSchema in packages/spec (the grep returns only FormFieldBaseSchema, MetadataTypeRegistryEntryBaseSchema and ObjectPermissionBaseSchema); BaseSchema.ariaLabel is objectui's OWN node interface in packages/types (pinned at packages/types/src/__tests__/base-schema-label-vocabulary.test.ts), and resolveAriaProps is objectui packages/react/src/SchemaRenderer.tsx:122. A metadata author cannot write BaseSchema.ariaLabel anywhere in this protocol, so that sentence would have been plausible and false in published prose. WHAT SHIPPED INSTEAD, measured: the tombstone names (i) the sibling 'description' on the same chart config, which the chart renderer lowers onto the chart graphic as role=img plus aria-label (liveness/dashboard.json records the objectui DOM pin 'announces chartConfig.description as the chart graphic accessible name'), and (ii) the three words ariaLabel / ariaDescribedBy / role where they really are authorable in this protocol - inside the shared AriaProps block on page.aria, page.components[].aria and the list view aria, the exact three surfaces the family's existing pin asserts are live. It names nothing retired: no app.aria, no dashboard.widgets[].aria as a destination. EVERY SHIPPED SENTENCE IN THIS DIFF: the retiredKey() guidance on ChartConfig.aria; two guidance entries for accessibility and ariaProps; the conversion's surface and summary; the extension of step18's rationale; the two retired-key entry-file comments; the changeset body. All of them regenerate into content/docs/references/ui/chart.mdx and report.mdx, reviewed in the diff. One shipped claim was WITHDRAWN before it landed: a first draft of the schema comment said AriaPropsSchema 'has seven other carriers'; the real count is 25 declaration sites, so the number was replaced with an unnumbered enumeration rather than shipping a count that rots.",
  "tests": "All exit codes captured by redirect-then-dollar-question, never through a pipe. BUILD: 'pnpm --filter @objectstack/spec build' under scripts/pm/os-verify-lock.sh gave 'VERDICT command-exit 0'. The FIRST build was exit 1 and is the useful reading: the gate itself named exactly the two keys to register - '2 key(s) were tombstoned with no registered retirement: ui/ChartConfig:aria, ui/ReportChart:aria'. TYPECHECK+TEST: 'pnpm --filter @objectstack/spec typecheck && pnpm --filter @objectstack/spec test' was exit 1 on the first run with exactly one failure - src/ui/chart.test.ts 'Chart ARIA Integration > should accept chart with ARIA attributes', a fixture pinning the acceptance this change removes, so it was INVERTED in place (there is no spelling of aria this shape accepts any more), with a positive control asserting description still parses. Re-run after the inversion: 'VERDICT command-exit 0', 'Test Files 481 passed (481)', 'Tests 13698 passed (13698)'. CHECK:GENERATED: exit 0 over 15 artifacts (2 proved stale and regenerated with --fix: content/docs/references and liveness/state-counts.md). The 14 source audits it names as NOT run were run as a group, all exit 0. check:objectui-pin-citations was a real red first (exit 1, 3 sites: the cited sha not written in backticks) and is exit 0 after the fix. check:i18n read exit 3 PREREQUISITE NOT MET first (NOT MEASURED, not green); its 10-package closure was built and it then read exit 0, '9 package(s) - all bundles in sync'. check:type-check-debt likewise moved 3 to 0 once the closure existed. LINT: 'pnpm lint' (repo-wide eslint . --no-inline-config) exit 0 - run in FULL, so NO narrowing is claimed and none is owed. GATE RECONCILIATION: 'node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack' derived 113 families; each was run with the capture idiom recording the command byte-for-byte beside its exit code; 'node scripts/pm/dispatch-gates.mjs --ran ran.list --repo objectstack-ai/objectstack' gave exit 0, '113 derived famil(ies) accounted for - 112 run, 1 NOT-MEASURED (1 DERIVED from a recorded exit 3), 0 UNRUN'. NOT MEASURED, declared: 'pnpm check:dual-build-cjs-loads', recorded exit 3 - it reads built output for studio, the connectors and several plugins, none of which this diff touches; that is a whole-monorepo build, which is CI's run. ABLATION - TWO LEGS, both mutated on disk, proved on disk, and restored by blob identity under a trap using absolute paths. LEG 1 (the tombstone): the mutation restored 'aria: AriaPropsSchema.optional()'. On-disk proof, anchored on the exact text removed: BEFORE aria-tombstone-lines=1 prescription-lines=1; AFTER aria-tombstone-lines=0 prescription-lines=0 live-declaration-lines=1; HEAD_BLOB=cbd0147784ead1076f2148671af25b9809172e34 and MUTATED_BLOB=2f422b6394e5a27f24ed18db3f2d3fd7c8d53bc2 differ, so the mutation reached disk. RED: ABLATED_PIN_EXIT=1, 'Tests 2 failed | 5 passed (7)', both failures carrying the pin's own message 'the tombstone must still REJECT - a passing parse means the key came back'. LEG 2 (the alias decision): the mutation restored the two alias entries and deleted the two guidance entries. BEFORE alias-entry-lines=0 guidance-entry-lines=1; AFTER alias-entry-lines=1 guidance-entry-lines=0; MUTATED_BLOB=cce7ea995b60582cf0f58dc765d1fcd5a1bfa716. RED: exit 1, 'Tests 2 failed | 24 passed (26)' - alias-integrity.test.ts naming both entries as pointing at a tombstone that 'accepts nothing', and the new pin failing with \"`accessibility` must carry the retirement: expected 'Unrecognized key(s) on this chart con...' to contain 'was removed in @objectstack/spec 17'\". RESTORE, proven by blob identity and NOT by an exit code (a cleanup step that does nothing also exits 0): after both legs RESTORED_BLOB=cbd0147784ead1076f2148671af25b9809172e34, equal to the HEAD blob of packages/spec/src/ui/chart.zod.ts, with 'git diff HEAD' empty and 'git status --porcelain' printing nothing. Both legs restored with 'git checkout HEAD --' against the path, never the bare 'git checkout --' form, which restores from the polluted index. GREEN AFTER: 7/7 in aria-carrier-tombstones.test.ts, 84/84 across it plus chart.test.ts. No ablation artefact remains in the tree. POST-REVIEW PUSH, re-measured on the NEW head: the last commit (16a8de837a) edits only a comment block in aria-carrier-tombstones.test.ts, and the ratchet family was re-run on that head rather than trusted from the earlier one. check:generated read exit 1 at first with check:api-surface stale - the stale-dist PHANTOM the gate names itself ('this gate reads the BUILT dist ... the removals above are phantoms'), because the comment edit moved the src digest. Rebuilt ('VERDICT command-exit 0') and re-read: check:generated exit 0, 'All 15 generated artifacts are up to date', working tree clean. Also green on the new head: check:type-check-coverage 0, check:cross-package-test-inputs 0, check:test-source-alias 0, check:tier-file-adoption 0, check:nul-bytes 0, the family pin 7/7, and repo-wide 'pnpm lint' 0 again.",
  "clause2_carrier_reading": "READING ONLY - I hung, cleared and waited on nothing, and called no label endpoint, so there is no endpoint or status code to report as refused. TWO readings, in order. AT PR OPEN: needs:contract-review was NOT on PR 18300 (labels read back as an empty list) and WAS on card 17751; 'node scripts/pm/check-clause2-carriers.mjs --pair 18300' exited 4, row C1 - the dual carrier hung on one side only, the expected shape for a PR the seat has not yet hung. AFTER THE SEAT'S STROKE, re-read at the end of this run: PR labels are [documentation, size/l, tests, tooling, needs:contract-review, protocol:ui], and '--pair 18300' now exits 0 - 'the clause-2 declaration is readable in the fixed spelling and both carriers agree.' The PR body carries the Clause-2 yes declaration line, which is a body line and not a label.",
  "mcp_calls": "0 - no MCP GitHub tool was called at any point in this run. Tool names called: (none). Every GitHub read and write went through the REST proxy with curl and the environment GITHUB_TOKEN.",
  "api_writes": "4 REST proxy writes + 7 git pushes. Endpoints: POST /repos/objectstack-ai/objectstack/pulls (1, the draft PR); PATCH /repos/objectstack-ai/objectstack/pulls/18300 (1, the body addition the seat asked for - see the footer note below); POST /repos/objectstack-ai/objectstack/issues/17751/comments (1, this report comment). git push to origin claude/issue-17751-chart-config-aria-retired (7: the empty-branch write-route probe plus six commits). NO labels endpoint was called at any point - the carrier is the seat's per the mid-run correction, and the budget slot it would have used is what the body PATCH spent. FOOTER HANDLING ON THE PATCH, done from the measured cell rather than by guess: platform-readings records that a bare REST PATCH of a PR body appends a bare footer while KEEPING the existing session-URL one, differing by exactly 58 bytes. So the PATCH deliberately sent a body with NO footer block and moved the durable session attribution into body prose, per AGENTS.md ('durable attribution lives in body prose or a comment'). Read back after the write: delta exactly 58 bytes, FOOTER_COUNT 1, session id present in prose, and no angle-bracket fragment in the stored body. The creation write was also read back: byte-identical to what was sent, one footer, session-URL form. The first push attempt was REFUSED by the pre-push hook (check:commit-card-trailers) because the harness-suggested co-author trailer named a model; AGENTS.md requires the model-free pair, so the commit was amended before anything was published.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: the CARD BODY item 4 asks for a major changeset and Clause-2 no. Both are stale against what this tree enforces - scripts/check-changeset-no-major.mjs refuses major during the launch window (breaking ships as minor until GA), and clause 2 covers accept-set movement in BOTH directions (packages/client/CHANGELOG.md records an 'Accept-set narrowing on a published SDK (clause-2)'). I followed the dispatch order and the gates: minor, Clause-2 yes. Surfaced here rather than silently chosen. No card edited - the delivering seat does not rewrite the card it was dispatched from. Successor: this card's reviewer, who reads the card beside the PR.",
    "noted, not filed: .claude/skills/spec-property-retirement/SKILL.md section 4 still instructs @objectstack/spec retirements to take a major changeset, which check:changeset-no-major refuses during the launch window. It is the playbook every new tombstone is authored from, so the next retirement card walks into the same contradiction. Governed surface (.claude), not this card's to edit, and it is a doc-vs-gate mismatch rather than a reproducible defect. Successor: the next retirement card through that playbook. Dedupe words: spec-property-retirement, changeset major, launch window, check-changeset-no-major, ADR-0049.",
    "noted, not filed: the same playbook's section 2 fork table reads 'strict schema means delete the key plus a guidance map', while every aria-family precedent on this tree is a retiredKey() tombstone INSIDE a strictObject - which shared/strict-object.ts documents as 'deliberate and strictly stronger than a guidance entry', and which aria-carrier-tombstones.test.ts relies on by name. The table is a default rather than a rule, but it points a reader the wrong way on exactly this class. Same governed surface and same successor as above. Dedupe words: strictObject retiredKey, fork table, acceptsNothing, tombstone strict, spec-property-retirement.",
    "noted, not filed: objectui at pin 53ded82bf7a4 authors chartConfig.aria in TWO of its own negative-pin TEST files (plugin-dashboard DatasetWidget.chartConfig.test.tsx and DatasetWidget.chartConfig.dom.test.tsx). No PRODUCTION objectui source does, and Console Pin Gate builds the Console SPA rather than running objectui's tests, so this PR does not red it - checked before opening. Those two pins are the consumer half that objectui#4044 is pm:blocked on by design, so this is already someone's card and filing it again would duplicate it. Carrier: objectui#4044."
  ],
  "deviations": "ONE scope question was put to me and here is the plain answer: ui/ReportChart:aria is (a) INSIDE the card's ruled scope, and it is not separable, so there is nothing for the seat to decide. The evidence, in the order it convinces: (1) ARITHMETIC, NOT A SECOND DECISION. Exactly ONE property on exactly ONE schema was retired - ChartConfigSchema.aria, the symbol the card names. ReportChartSchema is 'ChartConfigSchema.extend({ xAxis, yAxis })', so the extension COPIES the retired property into its own walked shape and authorable-surface marks the copy [RETIRED] separately. There is no second declaration to leave alone: the extension has no aria of its own to keep. (2) THE BUILD DEMANDED IT, UNPROMPTED. My first build after tombstoning the single property exited 1 on build-schemas.ts check (b) naming BOTH keys before I had written either entry: '2 key(s) were tombstoned with no registered retirement: ui/ChartConfig:aria, ui/ReportChart:aria'. The build stays red until both are registered, so 'retire ChartConfig.aria and leave ReportChart.aria alone' is not a state this tree can be in. (3) THE RULE IS WRITTEN DOWN. RETIRED_KEYS_BY_MAJOR's own docblock states this class on the shared/FieldMapping:transform precedent, where ONE tombstone produced THREE keys: 'every .extend() of the base copies the retired property into its own walked shape ... Registered per key, as the gate reads them - nothing radiates from the base.' (4) IT IS NOW VISIBLE IN THE PIN, as asked. packages/spec/src/ui/aria-carrier-tombstones.test.ts carries a dedicated case, 'the tombstone rides the .extend() onto ReportChart', and I added a paragraph above it saying in as many words that the second key is arithmetic rather than scope, that it is not separable, and that the build gate says so. Commit 16a8de837a. The PR body gained a matching section, 'Scope: why this registers TWO retired keys, and why that is not a widening'. WHAT THE THREE AUTHORED COORDINATES ARE, for completeness: the one declaration is authorable at dashboards[].widgets[].chartConfig.aria, reports[].chart.aria and reports[].blocks[].chart.aria, and the single D2 conversion strips all three. NO OTHER DEVIATION: no other ChartConfigSchema key moved, content/docs/releases was not touched, packages/spec/src/ui/stack.zod.ts was not touched (it does not exist - the file is packages/spec/src/stack.zod.ts, which I only READ, for the dashboards/reports collection declarations), and I hung, cleared and waited on no review label.",
  "docs_drift_reading": "No hand-written page is owed an edit, and I re-measured rather than taking it on trust. The advisory anchored exactly one page, content/docs/ui/dashboards.mdx, via the fixture string literal revenue_by_region and NOT via the retired key. Independently re-measured on origin/main 827cacbf4b (the tree, not the drift run's merge tree eacb3325ad): word-boundary aria in that page is 0 hits, grep -nE with the boundary form, exit 1; lit control in the same file and same idiom, chartConfig = 1 hit, which is the single row '| chartConfig | object | optional | Advanced chart configuration |' and does not enumerate the keys, so retiring aria falsifies nothing on it. The boundary form is load-bearing: a naive 'grep -i aria' returns two false positives in this very file, colorVariant and variable. NOT MEASURED, recorded as such: the same drift run reported 4 changed files yielding no anchor - one of them this PR's 18.ui__ChartConfig__aria.ts - and warned that its own checkout carried uncommitted changes, so its silence about pages documenting the retirement is NOT a clean bill of health and is not read as one. Both facts are in the PR body's acceptance notes."
}

---

Generated by Claude Code

[os-warren]

ACCEPT — #17751, PR #18300, head 16a8de837aec5791ba8976e80272f066bd818118

Contract review of record: #18300 (comment) — VERDICT: PASS, naming this head. Seat checklist below; the review's own reasoning is not restated here.

Tier, verified per transcript rather than taken on the reviewer's word

The reviewer's run transcript carries 173 occurrences of "model":"claude-fable-5-1" and zero of anything else, which is CONTRACT_REVIEW_TIER. The instrument discriminates: two os-dev transcripts from this same session, counted the same way, read 348 and 223 occurrences of "model":"claude-opus-5" and zero fable. So the at-tier claim is measured, not self-reported.

Landing pre-check, all three

1. At-tier PASS naming the current head — the record's Head-sha: is 16a8de837aec5791ba8976e80272f066bd818118, which is the PR's current head, read fresh. Read back from GitHub rather than from the reviewer's message: the five machine-read lines are stored intact, Served-tier: carries the constant's NAME, and the body holds no angle-bracket fragment.
2. node scripts/pm/check-clause2-carriers.mjs --pair 18300 → exit 0, and it confirms a review of record names this head.
3. Every check green: 42 check-runs on this head, total_count 42 equal to the 42 returned, 38 success, 4 skipped, 0 failing, 0 incomplete (15:37:56Z). The four skipped are Auto Label, Check PR Size, Console Pin Gate, Packed-tarball smoke (opt-in) — a subset of the set every one of this lane's three merged PRs this shift carried (fix(spec): guard four prototype fall-through lookups with own-property checks #18233 10 skipped, spec: closed duration types DurationMs / DurationSeconds beside EpochMs (step 1 of ruling A on #18115) #18238 7, fix(spec)!: the dateRange array arm is exactly two string bounds, and each refusal origin gets a true sentence #18230 7). A skip is not a pass, and this seat does not report them as green; the charter's line on this is itself unsatisfiable as written and is now filed as [finding] the enqueue bar 「每一个 check 全绿」 is unsatisfiable as written — every landing carries 7-10 skipped checks, and nothing distinguishes an expected skip from a path-filter miss #18308 for the skills seat.

Spot-check readings taken by this seat, not adopted from either agent

• Path surface, fresh at 15:37Z: 14 files. No .claude/**, no skills/**, no content/docs/releases/** — the PR is not on a governed surface and carries no release-owned edit.
• The fence held: packages/spec/src/shared/retired-key.ts is absent from the diff. It is spec: retiredKey() tombstones are emitted into a repeater row's items.properties, so the served schema offers six columns the parse door refuses #17502's declared surface and was fenced in the dispatch order.
• os-regen merge sequence: measured no-op, not skipped. The branch is behind origin/main by exactly one commit, 53dd5aaaaf, whose three files are .claude/settings.json, .claude/skills/pm-dispatch/SKILL.md and .claude/skills/pm-dispatch/references/rest-channel.md. Matched against the 18 merge=os-regen routes in .gitattributes: intersection NONE. The silent-drop hazard that sequence exists to prevent requires main to have advanced on an os-regen path the branch also edited; main advanced on zero of them.
• Report hygiene: mcp_calls is 0 with no tool names, so nothing in the deny roster was touched. api_writes is 4 REST endpoints, inside the budget.

Deviations carried forward

• The card body's item 4 (major changeset, Clause-②: no) is stale against this tree; the implementer shipped minor with Clause-②: yes, the reviewer confirmed scripts/check-changeset-no-major.mjs refuses major inside the launch window, and clause ② covers accept-set movement in both directions. Recorded, card not rewritten.
• ui/ReportChart:aria is arithmetic, not a widening: ReportChartSchema is a ChartConfigSchema.extend(...), and the build gate itself refused with both keys named before either entry existed. Now visible in aria-carrier-tombstones.test.ts.
• Two evidence-quality findings from the reviewer, neither changing the verdict: the ablation's recorded blob is from an earlier commit than the reviewed head (differing by one comment line, so the conclusion transfers but the reading was not stamped with its tree), and one prose phrase counting the aria family is false as a count while true as a lineage.
• For the maintainer batch, not decided here: .claude/skills/spec-property-retirement/SKILL.md contradicts this tree in two places the reviewer verified word for word — :291 directs @objectstack/spec retirements to a major changeset the launch-window gate refuses, and :87's fork table points away from the retiredKey()-inside-strictObject shape every aria-family precedent uses. It is the playbook every retirement card is authored from, so the next one walks into the same wall. Governed surface; this seat neither edits nor grades it.

Proceeding to clear both carriers, flip ready and enqueue.

---

Generated by Claude Code