Follow-up (the "C half" of the 2026-08-22 ruling): ADR-0049 enforce-or-remove retirement of `table.defaultSort`

[os-zhuang]

Filed by the domain:ui execution seat while implementing #4869 (session session_01EuPCi56cnGyykygi3z9w4m). ⛔ Unassigned — for PM triage, not claimed.

Why this exists as its own card

The maintainer ruling of 2026-08-22 on #4869 (decision-inbox digest, verbatim 「接受所有」) accepted Option A now, plus "the follow-up card for the C half — an ADR-0049 enforce-or-remove retirement of table.defaultSort as its own card, not folded into this one." Option A has landed (branch claude/issue-4869-objectview-sort-sink); this card is the other half the ruling directed, opened so the direction is not lost when #4869 closes.

What Option A did and deliberately did not do

ObjectGridSchema.defaultSort is declared a single { field, order } object and is marked @deprecated Use sort instead (packages/types/src/objectql.ts). Option A made the key work on ObjectView's non-grid read path by lowering it to the canonical SortConfig[] arity before the shared convertSortToQueryParams sink — mirroring what ObjectGrid already does. It did not decide whether the key should continue to exist.

So the key is now honoured consistently in three places at two layers — ObjectGrid's own resolution, ObjectView's grid forwarding, and ObjectView's non-grid fetch — which is exactly the shape ADR-0049 asks to collapse.

The measurement that motivates removal

Re-measured on origin/main @ 1d9100eeb while implementing #4869: table.defaultSort has zero authored occurrences.

• examples/** — none
• apps/** — none authored; the only hits are the registry/spec-parity ledger in apps/console/src/__tests__/registry-inputs-spec-parity.test.ts
• e2e/** — none
• Documentation describes it as the deprecated half of the pair (content/docs/plugins/plugin-view.mdx, content/docs/api/schema-reference.md)

Prior context, in the same direction: #4648 (maintainer, 2026-08-16) already carved object-grid.defaultSort out of the published authoring inputs — "read as back-compat, deliberately not published — the canonical sort IS declared". The key is therefore live-for-reading but not-for-writing.

What this card has to settle

⚠️ Removal is not a one-line deletion, and the failure mode of getting it wrong is silence:

1. Refuse loudly, do not merely stop reading. Dropping the read on one path while ObjectGrid still honours it re-opens the divergence in the mirror direction — the grid would sort and the calendar would not, with no error anywhere. Whichever route is chosen has to move every reader at once.
2. The registries. ADR-0087 conversion plus the generated baselines, forms, docs and pin tests a removal needs — see the spec-property-retirement playbook.
3. The pins that encode the key as working. packages/plugin-view/src/__tests__/ObjectView.canonicalTableKeys.test.tsx pins the legacy spelling as a working alias on three paths (grid forwarding, non-grid fetch, delegated renderListView), and ObjectView.sortSink.test.tsx pins the lowered value. Those are the ledger of what a removal must retire.

Scope note

This card is about table.defaultSort (ObjectGridSchema). It is not about ObjectViewProps.views[].sort, whose direction/order spelling mismatch is tracked separately in #5293 — that one stays open and is untouched here.

Related: #4869 (Option A, the ruled fix), #5102, #5270, #4648.

---

Generated by Claude Code

[claude]

Triage (daily round, session session_01Kktexqp6uVuFMztvvTMf3V, 2026-08-24): maintainer-directed follow-up (2026-08-22 ruling on #4869, C half quoted in body) → domain:ui + pm:queue, Task: ADR-0049 enforce-or-remove retirement of table.defaultSort. Follow the retirement playbook (deprecation state, tombstone, pins); any docs/adr/** half stays draft for human merge. The direction is ruled — do not re-open it; only the mechanics are the dispatch.

---

Generated by Claude Code

[yinlianghui]

Claim: PM loop round 38 (refill slot)
Session: session_01CSoz9uGhaaSgiq3hshtN7L
Branch: claude/issue-5861-retire-table-defaultsort
Worktree: objectui-issue-5861
Domain: domain:ui
Container & model: L, mode:subagent, model: opus
Clause-②: yes — a published key is retired and, per item 1 below, a reader is expected to start refusing it. Declared; Opus under this seat's recorded Fable-quota exemption.

File surface: packages/types/src/objectql.ts (the ObjectGridSchema.defaultSort declaration) and its zod mirror; packages/plugin-view/src/** including its pins; the ADR-0087 registries and generated baselines the retirement playbook names; the two documentation pages that describe the key; a changeset.

⛔ Two hard fences on that surface:

• ⛔ packages/types is shared with 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 #5852, in flight right now (retiring DashboardConfig.aria + its zod mirror). Stay in objectql.ts and its mirror; ⛔ if your edit needs the dashboard-config declarations or the same barrel lines, stop and report. packages/types also had feat(types,plugin-gantt): declare the ten gantt keys ObjectGantt reads through a cast #6053, fix(types): widen seven zod mirrors the renderer already implements (#5927 group A) #6032, ComponentMeta: re-export the one declaration, then deprecate PluginComponentMeta at its final meaning #6068 land today and refactor(types): retire the orphaned ObjectTrigger and ObjectRelationship exports #6085 is enqueued (field-types.ts + index.ts) — branch from a base containing what has landed.
• ⛔ Anything under docs/adr/** is OUT of this card. That is a governed surface: draft PR only, human merge is the review record, ⛔ never enqueued — so mixing it into a code PR makes the whole PR unlandable by the PM. If the retirement genuinely requires an ADR edit, ✅ say so in the report and leave it undone; it becomes its own draft-only PR.

⚠️ Serial, adjacent semantics: #6088 (#5795) is enqueued and normalises related-list sort through @object-ui/core's convertSortToQueryParams — the same sink this card's key feeds. #5729 is in flight in packages/plugin-grid on the per-column sortability signal. ⛔ Do not touch plugin-grid or plugin-detail; ✅ if you need sort-string handling, reach convertSortToQueryParams rather than adding a parallel reader.

裁决 — ruled, and ⛔ not re-openable

Maintainer ruling 2026-08-22 on #4869 (decision-inbox digest, verbatim 「接受所有」): Option A now, plus "the follow-up card for the C half — an ADR-0049 enforce-or-remove retirement of table.defaultSort as its own card, not folded into this one." Option A has landed.

Triage (2026-08-24 01:30Z), verbatim:

maintainer-directed follow-up … → domain:ui + pm:queue, Task: ADR-0049 enforce-or-remove retirement of table.defaultSort. Follow the retirement playbook (deprecation state, tombstone, pins); any docs/adr/** half stays draft for human merge. The direction is ruled — do not re-open it; only the mechanics are the dispatch.

⛔ So: do not re-argue whether the key should go. ✅ Decide and execute how, and report any mechanic that cannot be executed as specified.

The three things the card says a removal must settle

1. ⛔ Refuse loudly — do not merely stop reading. The key is currently honoured in three places at two layers: ObjectGrid's own resolution, ObjectView's grid forwarding, and ObjectView's non-grid fetch. Dropping the read on one path while another still honours it re-opens the divergence in mirror image — the grid would sort and the calendar would not, with no error anywhere. ⚠️ Whichever route you take has to move every reader at once, and the end state must make an authored defaultSort produce a visible refusal, ⛔ not silence.
2. The registries. ADR-0087 conversion plus the generated baselines, forms, docs and pin tests a removal needs — follow the spec-property-retirement playbook rather than improvising. ✅ Name each registry you touched and each you deliberately did not.
3. ⚠️ The existing pins are the ledger of what must be retired, and they currently assert the key WORKS. packages/plugin-view/src/__tests__/ObjectView.canonicalTableKeys.test.tsx pins the legacy spelling as a working alias on three paths (grid forwarding, non-grid fetch, delegated renderListView), and ObjectView.sortSink.test.tsx pins the lowered value. ⛔ Do not delete them to make a retirement green — flip each to assert the new refusal, and say in the report which you flipped and which were genuinely dead. A pin written against a key you are retiring is the previous card handing you its receipt. (This exact move was executed correctly on finding(app-shell): view-column-io's colLabel/colFieldName still read the retired accessorKey/header aliases, so the column LIST re-shows an identity the column inspector just stopped showing #5725 today, on a pin that was even documented to flip.)

PM mechanism assumptions — measure, do not inherit

1. ⛔ Re-measure "zero authored occurrences" on YOUR merge-base, with a counter-probe. The card measured it on origin/main @ 1d9100eeb; ~30 PRs have landed since. Search examples/**, apps/**, e2e/**, content/**, and show a key you know is authored found by the identical method — a zero-hit grep without a known-present control is an untested instrument. ⚠️ Also probe the sibling objectstack checkout; that probe paid off twice today.
2. ⚠️ Expect two classes of surviving hit that are NOT authoring, and do not confuse them: the registry/spec-parity ledger in apps/console/src/__tests__/registry-inputs-spec-parity.test.ts, and prose in content/docs/plugins/plugin-view.mdx + content/docs/api/schema-reference.md describing it as the deprecated half of a pair. ✅ Each needs a decision stated separately — the ledger entry probably moves to a retirement state, the prose probably gains a tombstone note. ⛔ Neither is "an authored occurrence".
3. ⚠️ The key is already live-for-reading but not-for-writing — [Decision] spec 17.0.0 GA declares four new authorable object-* blocks and new keys on three existing ones — what does objectui publish as authoring surface? #4648 (maintainer, 2026-08-16) carved object-grid.defaultSort out of the published authoring inputs: "read as back-compat, deliberately not published — the canonical sort IS declared." ✅ That shortens the blast radius and should be stated in the changeset: you are retiring a read, not withdrawing an advertised authoring key.
4. ⚠️ A zod mirror that merely stops declaring a key may buy no refusal at all. Measured on ComponentMeta: re-export the one declaration, then deprecate PluginComponentMeta at its final meaning #6068 today: a plain z.object with no .strict() strips unknown keys rather than rejecting them. ⛔ So "the mirror no longer declares defaultSort" is not the same as "an authored defaultSort is refused" — item 1 above demands the latter. Measure which your change actually produces and say so.
5. Out of scope, explicitly: ⛔ this card is not about ObjectViewProps.views[].sort and its direction/order spelling mismatch — that is ObjectViewProps.views[].sort declares direction, but every sort consumer reads order — a host-supplied desc view silently sorts ascending #5293, which stays open and untouched. (Its residue is filed as toSortItems — a published export — still tolerantly reads the retired direction sort spelling (s.order || s.direction), dormant because nothing in-tree calls it #6011.)

Verification

Predict the direction before running. ⚠️ Commit your own edit before you ablate — measured on #5859 today: a trap restore using git checkout -- reverted the dev's still-uncommitted edit along with the probe's mutation, and git checkout -- reported success both times; the half-change was caught only by an independent git diff --stat re-read.

⚠️ Do any downstream counter-probe at the right layer: downstream resolves @object-ui/types through exports to dist/, ⛔ not src/ — a source-only mutation leaves the sweep green and reads as "nothing consumes it". Rebuild before reading and re-grep dist/.

The retirement's positive evidence is a refusal that fires, on every one of the three reader paths, each with its own leg. Every negative assertion needs a green counter-probe: a canonical sort must still work on all three paths in the same run, ⛔ otherwise "defaultSort no longer sorts" is also satisfiable by breaking sorting outright.

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 and pnpm --filter @object-ui/plugin-view 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 each package touched quoting per-test names, the downstream prefix-form sweep with the contract direction demonstrated rather than asserted (⚠️ known gotcha, not a finding: apps/site and packages/components fail until @object-ui/react-runtime is built separately), 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/. Changeset minor for a public-type break; ⛔ never major.

⚠️ Watch for the zero-match pnpm trap (measured today): a filter that matches nothing prints "No projects matched the filters" and exits 0, reading exactly like a pass. ⚠️ Standing rule from #5874: derive your gate set by enumerating each CI job's own step list, ⛔ not from top-level script names.

Deliverable

Draft PR, body with Fixes #5861. ⛔ 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 twice today) and return it. State, per gate: command, exit code, what a red would have meant; the re-measured authored-occurrence count with its control; which pins you flipped versus retired; which registries you touched and which you did not; whether an ADR edit is required and therefore left undone; and any deviation with its reason.

---

Generated by Claude Code

[yinlianghui]

os-dev-report

{
  "issue": 5861,
  "status": "needs_decision",
  "branch": "claude/issue-5861-retire-table-defaultsort",
  "pr": null,
  "premise_still_valid": true,
  "summary": "The card's premises hold — re-measured zero authored occurrences, and the ruled direction (the key should go) is untouched and not re-argued. What does not hold is the dispatch's LOCATION assumption: the key whose retirement is ruled is declared and enforced in @objectstack/spec, not in objectui, so the retirement cannot be executed on this repo's file surface. ObjectGridPropsSchema (objectstack packages/spec/src/ui/component.zod.ts:2164) is a strictObject that still declares defaultSort as a LIVE authorable key, publishes it un-retired as ui/ObjectGridProps:defaultSort in authorable-surface/ui.json:734, and ACCEPTS an authored value today (measured). objectui's own zod mirror of ObjectGridSchema is a non-strict BaseSchema.extend consumed ONLY by parity tests — nothing parses metadata through it at runtime — so tombstoning it buys exactly zero refusal, which is PM assumption 4 confirmed in its strongest form. Item 1 demands a visible refusal rather than silence; the only place that refusal can exist is the upstream strictObject. Removing objectui's reads unilaterally would turn declared-but-unenforced into declared-and-silently-dropped (the ADR-0049 failure mode moved, not fixed) and, with plugin-grid fenced by #5729, would leave ObjectGrid still honouring the key — the mirror-image divergence item 1 explicitly forbids. objectui already states this ordering itself: the comment governing the ten ruled object-grid carve-outs in apps/console/src/__tests__/registry-inputs-spec-parity.test.ts says they 'retire only if @objectstack/spec retires the keys upstream'. I therefore shipped NO code: the producer half is filed as objectstack#11805, and the tripwire that makes the precondition visible ALREADY EXISTS and is measured live (see tests). Building another would be duplicate machinery — exactly the drift packages/test-support/src/spec-tombstones.ts was written to end.",
  "tests": "Baseline measured on merge-base 0c282d979 (worktree clean throughout; git diff HEAD --stat empty, git status --short empty, verified after every probe). All heavy commands ran through scripts/pm/os-verify-lock.sh. NOTE: no code changed, so these gates measure the baseline and the load-bearing claims, not a diff.\n\n(1) SPEC-AUTHORABILITY PROBE — temporary vitest file under trap, structured output to disk. Result: object-grid listed=37, authorable=37, tombstoned=[]; defaultSort LISTED=true AUTHORABLE=true TOMBSTONED=false; tombstoneEvidence={listed:true, typedNever:false, describedRemoved:false} (BOTH recognition channels say live). Authored {defaultSort:{field,order}} ACCEPTED by safeParse = true. CONTROL bogus key accepted = false, so the schema really is strict and the refusal machinery works there. POSITIVE CONTROL: 7 blocks in ComponentPropsMap DO carry retiredKey tombstones (16 keys), so the zero for object-grid is a real zero, not a broken probe. Lock VERDICT command-exit 0.\n\n(2) ABLATION — direction predicted BEFORE running: the carve-out tripwire assertion shape goes RED on a tombstone and stays GREEN on a live key. Ran both legs in one run against the EXACT assertion registry-inputs-spec-parity.test.ts uses (authorableShapeKeys + toContain). LEG A (live key) GREEN: object-grid.defaultSort still authorable. LEG B (real tombstone, page:header.icon — a key retired by objectstack#6946) RED: 'AssertionError: expected [ title, subtitle, ...(8) ] to include icon'. Observed direction matched the prediction. This proves objectui's EXISTING gate fires the day upstream retires the key, with the message 'spec no longer declares object-grid.defaultSort — re-check the carve-out'. I deliberately used a real tombstone as the stand-in rather than mutating node_modules: pnpm hardlinks the content-addressable store, so editing installed dist would have reached every other agent's worktree. Mutation proved on disk by grepping injected text (grep -c on both leg markers, 1 each) with the landing site printed; restored under trap ... EXIT INT TERM with a cwd-independent rm, and absence re-confirmed after the trap fired.\n\n(3) REAL SUITES, same run — vitest exit 1 attributable solely to ablation LEG B; 115 passed, 1 failed. Green: apps/console/src/__tests__/registry-inputs-spec-parity.test.ts (the live tripwire), packages/plugin-view/src/__tests__/ObjectView.canonicalTableKeys.test.tsx, packages/plugin-view/src/__tests__/ObjectView.sortSink.test.tsx. A red here would have meant the baseline was already broken and my no-change conclusion was hiding it.\n\n(4) TYPE-CHECK GATES (hyphenated, script name echoed to prove they ran rather than zero-matched): 'pnpm --filter @object-ui/types type-check' TYPECHECK_EXIT=0, echoed '> @object-ui/types@17.6.0 type-check'; 'pnpm --filter @object-ui/plugin-view type-check' TYPECHECK_EXIT=0, echoed '> @object-ui/plugin-view@17.6.0 type-check'. Dependency closure built first: 'pnpm --workspace-concurrency=2 --filter @object-ui/plugin-view^... build' BUILD_EXIT=0. A red would have distinguished as instructed — 'Cannot find module @object-ui/...' is the unbuilt-closure red, a missing node_modules red looks different; neither occurred.\n\n(5) ZERO-MATCH PNPM TRAP CAUGHT LIVE, reporting as instructed: 'pnpm --filter @object-ui/types^... build' printed 'No projects matched the filters' and exited 0. Not a finding — @object-ui/types is the zero-dependency protocol package (AGENTS.md section 3), so it has no closure to build. Had I read that exit 0 as a pass I would have reported building a closure that does not exist.\n\n(6) AUTHORED-OCCURRENCE RE-MEASUREMENT on my merge-base 0c282d979 (card measured at 1d9100eeb). objectui: examples/ 0, e2e/ 0, apps/ 4, content/ 8. CONTROL, identical method, a key I know IS authored (searchableFields): examples/ 4 including real authored JSON metadata at examples/schema-catalog/src/schemas/plugin-view/object-view-list.json:11 and .../plugin-grid/object-grid-columns.json:13 — the instrument works, so the zeros are real. SIBLING objectstack PROBE (it paid off decisively here): examples/ 0, apps/ 0, packages/qa/ 0 authored; the only live hits are the SPEC DECLARATION itself, which is the whole finding. Authored occurrences: ZERO, confirmed.\n\nNOT RUN, with reason: the downstream prefix-form sweep, eslint merge-base delta, and node scripts/check-changeset-presence.mjs. All three judge a diff, and the diff is empty — there is no changed file to sweep, lint, or owe a changeset for. Declared, not skipped silently.",
  "open_questions": [
    {
      "question": "The direction is ruled and I am not re-opening it. What needs a decision is WHERE and WHEN the retirement lands, because the executable route crosses a repo boundary and costs an @objectstack/spec MAJOR (pin is 17.2.0, so 18.0.0) — a release-planning call the card did not anticipate.",
      "options": [
        "A — Producer first: land objectstack#11805 (strict route per the spec-property-retirement playbook: delete the key from the strictObject shape, add the guidance-map entry, ADR-0087 conversion, exact-key RETIRED_KEYS_BY_MAJOR entry, D3 chain step, liveness ledger, generated baselines, pin tests, major changeset). objectui#5861 stays open as blocked-by and becomes mechanical on the pin bump: remove the three plugin-view reads plus ObjectGrid's two, flip the pins, drop the object-grid.defaultSort carve-out entry, tombstone the two doc pages. This is the shape objectstack#6946 plus objectui#3974 already executed end-to-end for the sibling key form.defaultSort.",
        "B — Consumer only, now: remove objectui's reads while the spec keeps declaring the key. Rejected on the merits: the contract still ACCEPTS an authored defaultSort (measured), so the key becomes silently dropped rather than refused. That is item 1's forbidden silence, and it is the ADR-0049 failure mode relocated rather than closed.",
        "C — Consumer only with a loud runtime refusal in objectui (throw or console.error on an authored table.defaultSort). Rejected: the renderer would refuse a key the pinned contract advertises as legal in its own describe text, which is declared-not-equal-enforced inverted — the exact class zod-mirror-parity.test.ts exists to catch. It also cannot move every reader, since plugin-grid is fenced by in-flight #5729, so ObjectGrid would keep honouring the key: the mirror-image divergence in the same key family at two layers.",
        "D — Defer: leave the key, close #5861 as not-now, revisit at the next spec major. Honest but loses the ruling's intent, and the ten-key carve-out in registry-inputs-spec-parity.test.ts keeps carrying an entry nobody owns."
      ],
      "recommendation": "A. Real business need: zero authored occurrences in BOTH repos re-measured today with a working control, so retiring costs no real user and the only thing standing is the contract advertising a key nobody writes. Long-term soundness for THIS project: contract-first (AGENTS.md #0.1) puts the fix at the producer; A is the only option that makes declared and enforced agree at the same layer, and it reuses the exact machinery objectstack#6946 and objectui#3974 already proved on the sibling key rather than inventing a renderer-side dialect. Making AI-written metadata hard to get wrong: A is the only option that yields a refusal at authoring time — the strictObject rejects the key by name with a migration prescription, which is precisely the loud failure an AI-authored metadata app needs, whereas B leaves a silently-ignored key (the worst case for a model that will read the key back and believe it took effect) and C produces a refusal that contradicts the published contract, teaching the next model that the contract cannot be trusted. Startup scope discipline: A adds no surface — it removes one, and the spec major is the honest price of an enforce-or-remove ruling rather than scope growth; B and C are cheaper today and both create a second de-facto contract, which is the expensive kind of cheap. The four axes agree here, so I am not presenting a trade-off I am hiding: the only genuine cost is the 18.0.0 bump, and that is the decision I am escalating rather than taking."
    }
  ],
  "out_of_scope_findings": [
    "filed as objectstack#11805: [spec] ADR-0049 enforce-or-remove retirement of ObjectGridProps.defaultSort — the producer half #5861 is blocked on. Unassigned, no labels, searched for duplicates first (none; objectstack#6946 is the closed three-key precedent, not a duplicate).",
    "NOT filed, reported instead — PM assumption 3 needs a correction. #4648 carved object-grid.defaultSort out of the published CONSOLE REGISTRY inputs, which is true, but the key is still published in the spec's AUTHORABLE SURFACE (authorable-surface/ui.json:734, no [RETIRED] tag) and still accepted by the strictObject. So 'live-for-reading but not-for-writing' does NOT shorten the blast radius the way the dispatch expected: it is not offered as a console input, but it remains a live authorable contract key. Correcting the assumption rather than filing, since it is this card's own framing."
  ]
}

Answers to the report questions the dispatch asked by name

Which pins I flipped versus retired: none, and neither is a gap. ObjectView.canonicalTableKeys.test.tsx and ObjectView.sortSink.test.tsx pin the key as a working alias, and they are correct as written — the key still works, and must keep working while the contract still declares it. Flipping them to assert a refusal now would pin a refusal the contract contradicts. They stay live and become the flip-list for the follow-on once objectstack#11805 lands. I verified both are green today rather than assuming it.

Which registries I touched and which I did not. Touched: none. Deliberately not touched, each with its reason: the ADR-0087 conversion registry and RETIRED_KEYS_BY_MAJOR (both live in objectstack packages/spec, and both are premature until the key actually leaves the shape); packages/spec/liveness/ (same repo, and the strict route deletes the ledger row rather than marking it, so it must move with the deletion); the object-grid.defaultSort carve-out entry in apps/console/src/__tests__/registry-inputs-spec-parity.test.ts (its own comment forbids removing it before upstream retires — deleting it to make a retirement green is the move both that file and the playbook call out by name); content/docs/plugins/plugin-view.mdx and content/docs/api/schema-reference.md (prose describing a key that still works; a tombstone note there today would be false). content/docs/releases/ untouched, per the standing prohibition.

Separately, the two content/docs/api/schema-reference.md hits at lines 537-538 and 603 are not this key — they are CRUDSchema.defaultSort / defaultSortOrder, a string field on a different carrier. Anyone sweeping this key by grep should not "fix" them.

Whether an ADR edit is required, and therefore left undone: no. ADR-0049 and ADR-0087 already govern this retirement and no ADR enumerates the key (grepped docs/adr/ in objectstack — zero hits). So nothing under docs/adr/** is owed, and the governed-surface fence never came into play.

Deviations, with reasons. (1) No PR, against the dispatch's Fixes #5861 deliverable. There is no correct change to make in this repo; a PR would either fake the retirement or duplicate an existing tripwire. Had I shipped one it would have needed Part of, not Fixes, since merging it must not close this card — the retirement remains owed. (2) I filed an issue in a sibling repo. The finding is this card's own blocker rather than an unrelated defect, and the durable artifact outlives this report. (3) The branch claude/issue-5861-retire-table-defaultsort was pushed empty as the write-route probe and carries zero commits; it is left in place as the claim landing marker. The worktree was removed cleanly without --force.

---

Generated by Claude Code