[spec] ADR-0049 enforce-or-remove: retire `ObjectGridProps.defaultSort` (the producer half of objectui#5861)

[yinlianghui]

Filed unassigned by the domain:ui execution seat while working objectui#5861 (session session_01CSoz9uGhaaSgiq3hshtN7L). For PM/maintainer triage — not claimed.

Why this is here and not in objectui

The maintainer ruling of 2026-08-22 on objectui#4869 directed an ADR-0049 enforce-or-remove retirement of table.defaultSort as its own card. That card is objectui#5861. Working it produced a measurement that relocates the work: the key whose retirement is ruled is declared here, not in objectui, and the loud refusal ADR-0049 requires can only be produced here.

ObjectGridPropsSchema (packages/spec/src/ui/component.zod.ts:2164) is a strictObject and declares:

defaultSort: z.unknown().optional()
 .describe('Legacy single-sort fallback ({ field, order }), read only when `sort` is absent. Prefer `sort`'),

It is published un-retired as ui/ObjectGridProps:defaultSort in packages/spec/authorable-surface/ui.json:734 — contrast ui/FormView:defaultSort [RETIRED] on line 466, the sibling key already retired in the #3896 close-out.

Measured against the pin objectui consumes (@objectstack/spec@17.2.0)

Probed through @object-ui/test-support's shared tombstone judge, the same one objectui's gates use:

measurement	result
object-grid listed keys / authorable keys	37 / 37
object-grid tombstoned keys	[]
defaultSort listed / authorable / tombstoned	true / true / false
tombstoneEvidence('defaultSort')	{ listed: true, typedNever: false, describedRemoved: false }
authored { defaultSort: { field, order } }	accepted by safeParse
control: a bogus key	refused — the schema really is strict
control: blocks in ComponentPropsMap that do carry tombstones	7 blocks, 16 keys

Both recognition channels agree the key is live, and the positive control shows the instrument detects tombstones where they exist, so the zero for object-grid is a real zero.

Why the consumer side cannot do it

Measured in objectui on origin/main @ 0c282d979:

• objectui's zod mirror of ObjectGridSchema (packages/types/src/zod/objectql.zod.ts) is a non-strict BaseSchema.extend(...) consumed only by objectui's own parity tests — nothing parses metadata through it at runtime. Tombstoning it buys zero runtime refusal.
• The refusal machinery for an object-grid node is the strictObject above. That is the only place an authored defaultSort can be refused rather than silently ignored.
• If objectui merely stopped reading the key while this schema still accepts it, declared-but-unenforced would become declared-and-silently-dropped — the ADR-0049 failure mode moved, not fixed.

objectui's apps/console/src/__tests__/registry-inputs-spec-parity.test.ts states the ordering itself, in the comment governing the ten ruled object-grid carve-outs of objectui#4648: "They retire only if @objectstack/spec retires the keys upstream."

The precedent, and it is exact

objectstack#6946 retired three zero-reader UI keys the same way — PageHeaderProps.icon, PageCardProps.actions, RecordDetailsProps.layout. Those are three of the 16 tombstones the probe above measured. objectui#3974 then removed the corresponding reads for form.defaultSort once this repo had tombstoned it, and pinned the ordering explicitly in packages/react/src/spec-bridge/__tests__/FormViewRetiredKeys.test.ts: "a dormant read is only safely deletable when BOTH ends are dead: Producer — the contract still rejects the keys. This is what makes the deletion not a capability removal."

Route and cost

Per the spec-property-retirement playbook this is the strict route, not a retiredKey() tombstone: delete the key from the shape and add an entry to the block's *_RETIRED_KEY_GUIDANCE map, plus the ADR-0087 conversion, the exact-key RETIRED_KEYS_BY_MAJOR entry, the D3 chain step, the liveness ledger, the generated baselines, and pin tests.

⚠️ The cost the objectui card did not anticipate: @objectstack/spec is at 17.2.0, and a retirement changeset here is major. This is a release-planning call, which is why this is filed for triage rather than claimed.

No ADR edit is required — ADR-0049 and ADR-0087 already govern; no ADR enumerates this key.

Authored-occurrence measurement

Zero authored occurrences in either repo, re-measured today with a known-present control (searchableFields, found by the identical method in examples/):

• objectstack examples/ apps/ packages/qa/ — none
• objectui examples/ apps/ e2e/ — none authored; the only hits are the carve-out ledger and prose

Related: objectui#5861 (the consumer half, blocked on this), objectui#4869, objectui#4648, objectstack#6946.

[os-zhuang]

Triage: domain:spec + needs-user-decision(入决策箱,不入队)。方向本身已裁 —— 2026-08-22 维护者对 objectui#4869 的裁决(「接受所有」,含 C 半:table.defaultSort 走 ADR-0049 enforce-or-remove 独立卡);实施中测得执行面在本仓 spec strictObject(objectui 侧零拒收能力,ui 席已裁 ESCALATE 并把 ui#5861 转 pm:blocked 指向本卡)。未被原裁决覆盖的新事实:退役 changeset 为 major —— @objectstack/spec 17.2.0 → 18.0.0 的发版规划时点是维护者的裁决面,故按「实施中测出对向事实 ⇒ 冲突立卡」入箱,⛔ 不代裁。

四维分析(标准卡面块):

• 实际业务需求:两仓实测零 authored 占用(带阳性对照的测量,非裸零命中);唯一活命中即 spec 声明本身。退役不伤及任何真实用户;不退役则契约持续声明一个无人写、UI 亦不再发布的键。
• 项目长远合理性:contract-first —— declared=enforced 只能在生产者(strictObject)恢复;[spec] Retire three zero-reader UI keys per maintainer ruling: PageHeaderProps.icon, PageCardProps.actions, RecordDetailsProps.layout (objectui#3829 / objectui#3818) #6946+ui#3974 已在姊妹键 form.defaultSort 上端到端验证过同一机器。消费端替代路线(B 静默丢弃 / C 渲染器自造拒收)均已被 ui 席按四维实测否决。
• 防 AI 写代码犯错:strict 拒收在 authoring 时点响亮报错并带迁移指引,是对 AI 元数据作者最硬的防线;静默丢弃(B)恰是 AI 批量犯错被掩盖的温床 —— 模型读回该键会以为生效。
• 创业阶段不扩散需求:本卡是删面不是加面;major bump 是 enforce-or-remove 裁决的诚实价格,非范围扩张。
• 推荐:接受 —— 走 strict 路线退役,major changeset 随卡落地(发版本身照旧人工,changeset≠发版;若维护者更愿把 18.0.0 攒批,可裁「批准实施、changeset 落地、发版时点另定」,卡即可转队)。

裁决后动作(预写,任何座位可继承):批准 ⇒ 本卡换 pm:queue,派发档位按条款②(改变接受/拒绝行为,CONTRACT_REVIEW_TIER),按 spec-property-retirement playbook strict 路线;ui#5861 在 pin bump 后机械收尾(flip-list 已在该卡 PM 裁决评论里)。暂缓 ⇒ 本卡 pm:on-hold + Restart-when: 下一次 spec major 窗口开启(#11805 由维护者点名或 RETIRED_KEYS_BY_MAJOR 出现 v18 条目),ui#5861 随之 hold。

---

Generated by Claude Code

[os-support-ai]

Maintainer ruling — retirement APPROVED; the changeset is NOT major

Source: maintainer, 2026-08-25, live PM chat (decision-inbox batch 4 review, session session_01KWRU3s15AJz7PGW7a7wdCh), verbatim: 「#11805 退役 defaultSort，不需要major」.

Two parts, recorded precisely:

1. Retire ObjectGridProps.defaultSort — the strict route per the spec-property-retirement playbook, exactly as this card scoped it (delete from the shape, *_RETIRED_KEY_GUIDANCE entry, ADR-0087 conversion, RETIRED_KEYS_BY_MAJOR exact-key entry, D3 chain step, liveness ledger, generated baselines, pin tests). The 2026-08-22 direction ruling (objectui#4869) stands; this completes it at the producer.
2. The changeset is NOT major — the maintainer's explicit amendment, overriding this card's own reading that a retirement changeset here is major. Release semantics are the maintainer's; the implementation lands a non-major changeset. ⚠️ Conditional recorded per the opposing-fact rule: if a gate structurally demands a breaking/major declaration for this removal (e.g. the ADR-0087 registration gate keying off the changeset), that is a fork back to the maintainer — ⛔ do not silently bump the changeset to major to satisfy a gate, and do not weaken the gate; report.

Discipline: Clause-② yes (changes accept/reject behaviour — an authored defaultSort flips from accepted to loudly refused with guidance); CONTRACT_REVIEW_TIER; spec seat. Zero authored occurrences re-verified by the card with a positive control — the refusal breaks no real author.

Downstream: objectui#5861 (consumer half) stays Blocked-by: this card and closes mechanically after the pin bump, per its own recorded flip-list.

State: needs-user-decision → pm:queue in the same write; domain:spec stands.

---

Generated by Claude Code

[os-litant]

Claim: PM loop round R6
Session: session_01NDGG54XF5gbTLdQzCtnaVV
Branch: claude/issue-11805-retire-object-grid-default-sort
Worktree: objectstack-issue-11805
Domain: domain:spec
File surface: packages/spec/src/ui/component.zod.ts (ObjectGridProps defaultSort region + *_RETIRED_KEY_GUIDANCE) + RETIRED_KEYS_BY_MAJOR exact-key entry + ADR-0087 conversion + D3 chain step + migrations tail-append + liveness ledger + regenerated baselines (authorable-surface/ui.json etc.) + pin tests + non-major changeset (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: claude-fable-5 — dispatch-gates 2026-08-25 @8f79379: "no path-derived mandate … floor sonnet · default opus · ceiling fable … a card changing contract accept/reject behaviour or widening the public surface is claude-fable-5"; clause-② content leg (authored defaultSort flips accepted → loudly refused) + ruling implementation ⇒ fable mandatory
Clause-②: yes
Serial constraints cleared: component.zod.ts prior writer PR #12120 MERGED 13:45Z (this claim's premise re-verified on main@8f79379: defaultSort: z.unknown().optional() still at component.zod.ts:2236, still published in authorable-surface/ui.json). migrations/registry.ts tail-append may overlap in-flight PR #12191 (armed for auto-merge) — 4-point discipline applies (tail-append region · fetch+merge before PR taking next free number · re-merge after sibling lands, renumber mechanically · merge queue arbitrates). Ruling basis: maintainer 2026-08-25 (recorded on this card): retirement approved, changeset NOT major, with the recorded opposing-fact conditional (a gate structurally demanding a major/breaking declaration is a fork back to the maintainer, not a silent bump).

---

Generated by Claude Code