keep ListViewSchema.fieldOrder — declare the columns × hiddenFields × fieldOrder composition in the contract, with pins, and re-cite the liveness row (ruling B, 2026-09-11; supersedes the 2026-09-04 retirement ruling whose premise was measured false)

Filed by the director seat from the maintainer's ruling on objectui#7516 (2026-09-04, decision batch #28, verbatim 「同意」 on recommendation A — retire at the spec). domain:spec set from the ruling's named executor; triage corrects if the landing point disagrees. Clause-②: yes (a published contract's accept set narrows) — contract-review tier, needs:contract-review on the PR.
✅ UNBLOCKED — the C9 blocker cleared when this seat took over the unreleased claim on the maintainer's word (takeover comment 5790817662, 2026-09-23T08:14Z); the card is pm:dispatched again.

What to retire

packages/spec/src/ui/view.zod.ts:1794 — fieldOrder: z.array(z.string()).optional().describe('Explicit field display order for this view') on the list-view schema (measured on origin/main e293038, 2026-09-04T05:03Z).

Why (the ruling's reasoning, for the implementer)

columns on the same schema is an ordered array and is the authoritative column order every host relays. fieldOrder is a second expression of the same thing, and no contract defines who wins when they disagree (objectui's ListView.tsx:2265 applies fieldOrder after columns, so it wins by implementation order, not by declaration). One slot, one spelling.
Authored instances: zero in objectstack outside packages/spec/src/ui/view.test.ts (two test documents) and the generated docs; zero in objectui packages/** / apps/** / examples/**. The hiddenFields control returns 14 files in objectstack, so the zero is a reading.
The liveness ledger (packages/spec/liveness/view.json:164) records fieldOrder as live on evidence that names objectui: packages/react/src/spec-bridge/bridges/list-view.ts:124 — a path that no longer exists on objectui origin/main 4bb5e10. The ledger row must be re-measured as part of this retirement, not trusted.

Route

Follow the spec-property-retirement playbook (ADR-0049 enforce-or-remove): named refusal / tombstone for the key, ADR-0087 conversion entry, the liveness ledger row, generated baselines (authorable-surface*.json), content/docs/references/ui/view.mdx regeneration, the two view.test.ts fixtures, and the objectstack lint rule packages/lint/src/validate-list-view-field-refs.ts (it validates fieldOrder references — it retires or narrows with the key; validate-react-page-props.ts and reference-integrity-suite.ts reference it too, re-derive on the tree). Changeset per the spec's BREAKING-as-minor convention with the migration text: "use the order of columns".

Stop conditions

A consumer census with a positive control before deleting: objectstack, objectui, and the sibling repos (hotcrm, cloud) for authored fieldOrder on list views. Any real authored instance ⇒ stop and report on objectui#7516; the ruling's fallback (keep, and define columns × fieldOrder × hiddenFields precedence with pins) is then re-presented.
⛔ Do not touch objectui's reader in this PR — objectui#7516 carries the reader removal and is Blocked-by: this card; it lands after the spec pin covering this retirement is consumable there.

Refs: objectui#7516 (ruling and facet block) · objectui#7218 (rowColor, the other half, unaffected) · ADR-0049 · ADR-0087.

Generated by Claude Code

Claim: session_01MkQhmuuJAVDjmeWNixwDDH — branch claude/issue-15184-retire-list-view-field-order

Clause-②: yes

Retiring an authorable key narrows a published accept set — the clause-② trigger by the mechanical boundary test, exactly as the card declares. ⇒ contract-review tier, and this seat commissions an at-tier review on the final head before anything lands.

⛔ Changeset: NOT asserted. Measure whether one is owed and apply the measured answer. ⚠️ Likely yes here — packages/spec/src/**/*.zod.ts is in that package's published files[] roster — but measure it; I have twice ordered a changeset the measurement refused.

Dispatched by the domain:spec execution seat at 2026-09-10T21:45Z. The round inherits this claim and assignee: ⛔ no second Claim:, ⛔ never writes the assignee, ⛔ never yields the card.

⚠️ READ THIS FIRST — the card one place ahead of you in this queue had a FALSE premise of the same shape

#15180 was dispatched today to retire an authorable key on the strength of "declared, accepted, read by nothing". Measured: the key was read, server-side, and retiring it would have deleted a live channel reaching the screen. The round stopped, correctly, and shipped nothing.

This card says "zero authored instances outside spec's own tests". ⭐ That is a claim about INSTANCES, ⛔ not about CODE THAT READS THE KEY. Those are different questions and #15180 turned on exactly that difference. Both must be answered before anything is removed.

Seat's pre-measurement — `origin/main` `6e3462df47`, 2026-09-10T21:43Z

The declaration (⚠️ the card's :1794 has drifted — anchor by CONTENT):

packages/spec/src/ui/view.zod.ts:2072
  fieldOrder: z.array(z.string()).optional().describe('Explicit field display order for this view'),

⛔ FOUR packages/lint source files read this key — the card does not mention them, and they are part of this work:

packages/lint/src/validate-list-view-field-refs.ts   :59 :154 :312   (stale-entry diagnostics)
packages/lint/src/validate-field-consumers.ts        :208            (in the field-bearing key list)
packages/lint/src/validate-react-page-props.ts       :708 :711       (in the fields: [...] list)
packages/lint/src/reference-integrity-suite.ts       :57 :251 :280   (reference-integrity coverage)
LIT CONTROL  'hiddenFields' across packages/ apps/ examples/ : 16 files — so the readings above are real

These validate that fieldOrder entries name real fields. A retirement must decide what happens to each, ⛔ rather than leave four validators policing a key nobody can author.

⭐ THE STOP CONDITION — the liveness ledger says this key is LIVE, on cross-repo evidence:

packages/spec/liveness/view.json  /props/list/children/fieldOrder
  "status": "live", "evidenceScope": "cross-repo",
  "evidence": "objectui: packages/react/src/spec-bridge/bridges/list-view.ts:124 copies the authored key
               onto the list node; objectui: packages/plugin-list/src/ListView.tsx:1499-1500 builds an
               order map from schema.fieldOrder and reor…"

The card asserts that first path no longer exists on objectui origin/main. ⚠️ I could not verify that from this repo — it is a cross-repo claim and objectui is not checked out here. ⇒ Falsify it before removing anything.

⛔ If either objectui consumer still exists and still reads schema.fieldOrder, STOP and report. That would make this #15180's shape a second time — a "dead" key that is in fact live one repo over — and the card becomes a decision, not a chore. A zero-implementation stop with evidence is a good outcome.

The ruling this executes

Maintainer ruled A — retire at the spec on objectui#7516 (2026-09-04, decision batch #28, verbatim 「同意」). ⛔ The round does not re-open whether to retire; it establishes whether the premise the ruling rested on still holds, and reports if it does not.

Decision frame: the canonical four in pm-dispatch 〈升级与决策〉 — 实际业务需求 / 项目长远合理性 / 防 AI 写元数据 app 犯错 / 创业阶段不扩散.

Generated by Claude Code

os-dev-report

{
  "issue": 15184,
  "status": "needs_decision",
  "branch": "claude/issue-15184-retire-list-view-field-order",
  "pr": null,
  "premise_still_valid": false,
  "summary": "STOP CONDITION MET — nothing implemented, no branch pushed, no PR opened. The dispatch's stop condition was: if either objectui consumer still exists and still reads `schema.fieldOrder`, stop and report. One of the two does. Measured on objectui origin/main 35c6a34532dce64de3369b77169c5993ebdf80c2. (1) Ledger evidence path A, `objectui: packages/react/src/spec-bridge/bridges/list-view.ts`, is GONE — REST contents HTTP 404 at ref=main, and the `spec-bridge/bridges/` directory does not exist in the tree. The card is correct about that path. (2) Ledger evidence path B, `objectui: packages/plugin-list/src/ListView.tsx`, is ALIVE, with only its line numbers drifted: the ledger cites :1499-1500 plus :1509, the code is now at :2424-2425 plus :2434 — it builds `new Map(schema.fieldOrder.map(...))` and sorts the effective column set by it, with `schema.fieldOrder` as a memo dependency. (3) The authored value REACHES that reader — the leg #15180 turned on: `packages/react/src/SchemaRenderer.tsx:1590` hands the authored node verbatim to the registry component as its `schema` prop; `ListView.tsx:882` runs it through `normalizeListViewSchema`, which is copy-on-write (`const next = { ...s }`) and strips no unrecognised key, so `fieldOrder` passes through untouched; the memo at ListView.tsx:2411-2434 then applies projection (`columns`), the FLS gate, the `hiddenFields` filter, and finally the `fieldOrder` sort. On the app-shell host path the same reader is fed by `ObjectView.tsx:2378`, whose `fullSchema` opens with `...listSchema` — the HOST's list schema — so an authored `fieldOrder` on the object's list schema reaches it there too. So `ui/ListView:fieldOrder` is live end-to-end, authored metadata to rendered column order, and retiring it deletes a channel that reaches the screen. The liveness ledger's `live` verdict at packages/spec/liveness/view.json /props/list/children/fieldOrder is CORRECT on re-measurement; only its evidence STRING is half-stale (dead path A, drifted line numbers on path B). The card's inference — the cited path is gone, therefore re-measure it dead — does not hold. This is #15180's shape a second time.",
  "tests": "NOT MEASURED — and not owed. Zero files were changed by this round, so no gate family is owed and none was run; `check:liveness`, `check:adr-0087-registration`, `check:authorable-surface`, `check:api-surface`, `check:docs` and `check:generated` are all NOT MEASURED because there is no diff to measure. No ablation was run, for the same reason: nothing was mutated. `scripts/pm/os-verify-lock.sh` was never taken. The round's product is measurement, and every reading below is a cross-repo static measurement, each with a lit control in the same command. CENSUS, objectstack origin/main, `examples/` + `apps/`, files matching, one command: columns 54, grouping 10, rowColor 1, hiddenFields 0, fieldOrder 0. The control is lit and the zero is a reading — but note what it reads: `hiddenFields`, which is indisputably live (it has its own relay rung at objectui ObjectView.tsx:2426 and its own filter at ListView.tsx:2418-2423), scores the SAME zero. So `zero authored instances in our own repos` is true for `fieldOrder` and is not evidence of deadness for this key family at all. CENSUS, objectui origin/main, `examples/` + `apps/`: hiddenFields 4 (lit control), fieldOrder 0 — a real reading, and it agrees with the card: nobody in objectui's own apps authors it either. CENSUS, objectstack origin/main, whole tree, 16 files carry `fieldOrder`, fully accounted for: the declaration (packages/spec/src/ui/view.zod.ts:2072 — the seat's re-anchored line, confirmed by content; the card's :1794 has drifted), two authored fixtures in packages/spec/src/ui/view.test.ts:2512 and :2595, four lint validators plus one lint test, two generated authorable-surface files carrying BOTH `ui/ListView:fieldOrder` and `ui/ObjectListView:fieldOrder`, the liveness row, four generated docs files, and two CHANGELOGs. There is NO objectstack runtime or server-side reader, which is correct and expected — column display order is a client concern — so the objectstack half of the census matches the card. FILE-EXISTENCE probes were run against the GitHub contents API, never by grepping contents for a name.",
  "mcp_calls": "0 — every GitHub read went through repo-scoped REST (the session's REST probe returned HTTP 200) or through git against the local objectui clone's fetched origin/main; no MCP GitHub call was made this round",
  "open_questions": [
    {
      "question": "objectui#7516's ruling A (retire at the spec) rested on the premise that `fieldOrder` is a second, effectless spelling of the order `columns` already carries. Measurement falsifies that premise: `columns` is the field PROJECTION, `hiddenFields` subtracts from it, and `fieldOrder` orders what survives (entries absent from `fieldOrder` sort to Infinity, i.e. to the end). The spec declares the latter two as a deliberate pair under one comment — `Field Visibility and Ordering per View (Airtable-style)`, view.zod.ts:2070-2072 — and objectui applies all three in that order in a single memo. What is genuinely undefined is the CONTRACT (nothing in the spec says how the three compose), not the BEHAVIOUR. The maintainer ruled on facts that do not hold; this round does not re-open whether to retire, it reports that the premise failed and hands the decision back. Which way now?",
      "options": [
        "A — execute the ruling as written and retire the key anyway. 实际业务需求: negative — it deletes a working channel; any customer app that authors `fieldOrder` on a list-view node today gets its column order applied, and after retirement gets a parse rejection and loses the ordering. 项目长远合理性: negative — ADR-0049 is enforce-or-remove, and this key is already ENFORCED (declared in spec, validated by four lint rules, accepted by objectui's type mirror by reference, read and applied by the renderer); removing an enforced key is not what ADR-0049 asks for. 防 AI 写元数据 app 犯错: mixed — it does collapse two spellings to one, but at the cost of a rejection where behaviour used to exist. 创业阶段不扩散: negative — it also lands a red pin one repo over: objectui's `ObjectView.relayRungCensus-7559.test.ts` requires `fieldOrder` to be a MEMBER of the zod mirror (:519, :762) and requires ListView to still HAVE a reader (:545, the `known-gap` kind's mechanical evidence), and that mirror imports the key BY REFERENCE from `SpecListViewSchema.shape` (objectui packages/types/src/zod/objectql.zod.ts) — so the retirement goes red in objectui at the next spec bump, while this card forbids touching objectui.",
        "B — the ruling's own stated fallback: KEEP the key and define the `columns` times `hiddenFields` times `fieldOrder` precedence in the spec contract, with pins. 实际业务需求: positive — nothing a user sees changes, and the composition authors already get becomes the one the contract promises. 项目长远合理性: positive — it converts a de-facto composition into a declared one, which is the ADR-0049 `enforce` branch this key actually qualifies for. 防 AI 写元数据 app 犯错: strongly positive — the real hazard today is not two spellings, it is that a model reading the spec cannot tell how they compose; writing the composition down is exactly the fix. 创业阶段不扩散: positive — no cross-repo bump, no red pin, no migration, and it retires the objectui#7516 question rather than deferring it.",
        "C — retire only the per-view spelling `ui/ObjectListView:fieldOrder` and keep the list-node spelling `ui/ListView:fieldOrder`. 实际业务需求: neutral-to-positive — the per-view spelling is the one that is silently dropped today (objectui's `renderListView` relay has no `fieldOrder` rung), so removing it costs nobody anything that works. 项目长远合理性: positive — it deletes exactly the half that is dead and keeps exactly the half that is live, which is the honest reading of the measurement. 防 AI 写元数据 app 犯错: strongly positive — an accepted key that is silently discarded is the precise trap class this axis exists for. 创业阶段不扩散: negative — this is NOT this card's decision to take: the per-view gap is objectui#7516 itself, which is OPEN and carries `needs-user-decision`, and its two options are `add the missing rung` versus `retire that surface`. Taking C here would pre-empt that card, and it is still a clause-2 narrowing needing its own review."
      ],
      "recommendation": "B, and re-present on objectui#7516 exactly as that card's own stop condition instructs. B is the ruling's own named fallback, it is what the measurement supports, and it is the only option that neither deletes a working channel nor pre-empts an open decision card. Two riders for whoever executes: (i) the `columns`/`hiddenFields`/`fieldOrder` composition should be written into the spec's `.describe()` text and pinned, since the composition is currently knowable only by reading objectui's ListView memo; (ii) C is the variant the maintainer may reasonably prefer, but it belongs to objectui#7516 and should be decided there, not here. Whichever is chosen, the liveness row's evidence STRING needs re-citing in the same PR: it names a deleted path and stale line numbers, and leaving it is what set this round up to nearly delete a live key."
    },
    {
      "question": "Changeset measurement, ordered by the dispatch and answered here for whoever executes the decision (moot this round, since nothing ships).",
      "options": [
        "Measured: `@objectstack/spec`'s published `files[]` roster is [dist, json-schema, liveness, prompts, llms.txt, README.md, src/**/*.zod.ts, CHANGELOG.md, api-surface, spec-changes.json] at version 17.4.0.",
        "Therefore BOTH candidate edit sites are published: `packages/spec/src/ui/view.zod.ts` matches `src/**/*.zod.ts`, and `packages/spec/liveness/view.json` sits under the published `liveness/` directory."
      ],
      "recommendation": "A changeset IS owed for any edit to either file — the seat's `likely yes` is confirmed by measurement, not assumed. Under option B or C that is a `minor` per the spec's BREAKING-as-minor convention for a narrowing; under a ledger-evidence-only correction it is still owed, because `liveness/` ships."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed: the liveness row `packages/spec/liveness/view.json` /props/list/children/fieldOrder carries half-stale evidence — its first citation `objectui: packages/react/src/spec-bridge/bridges/list-view.ts:124` names a file deleted from objectui, and its second citation's line numbers have drifted from :1499-1500/:1509 to :2424-2425/:2434. The VERDICT (`live`) survives re-measurement; only the string is wrong. Not filed because this card already commissions the re-measurement in its own Route section, so the correction belongs to #15184's disposition, whichever way it goes — not to a separate card. Successor: whoever executes the decision above. `verifiedAt` is 2026-08-10 and the row is not yet expired under the 180-day cross-repo rule, so no gate will catch this before roughly 2027-02-06.",
    "noted, not filed: the per-view spelling `ui/ObjectListView:fieldOrder` is declared, accepted, lint-validated and then SILENTLY DROPPED — objectui's `renderListView` relay (ObjectView.tsx:2377-2520) has a rung for its sibling `hiddenFields` at :2426 but none for `fieldOrder`. That is the metadata-trap class (a key an author can write that the runtime discards without a word). Not filed because it is already carried by objectui#7516, open with `needs-user-decision`, and already declared mechanically in objectui's own ABSENCES ledger at `packages/app-shell/src/views/ObjectView.relayRungCensus-7559.test.ts:480` as `kind: known-gap`. Successor: objectui#7516.",
    "noted, not filed: `fieldOrder` in objectui's `packages/app-shell/src/views/metadata-admin/` (SchemaForm.tsx, ResourceEditPage.tsx, registry.ts, default-schemas.ts) is a NAME COLLISION, not an instance of this key — it is metadata-admin's own form-field ordering prop, unrelated to `ListViewSchema.fieldOrder`. Recorded so the next census does not double-count it. Successor: none.",
    "noted, not filed: the four `packages/lint` validators the seat surfaced (validate-list-view-field-refs.ts:59/:154/:312-314, validate-field-consumers.ts:208, validate-react-page-props.ts:708/:711, reference-integrity-suite.ts:57/:251/:280) are all correct as they stand and need no change under options B or C — they police that `fieldOrder` entries name real fields, at `warning` severity for stale entries. Recorded because the dispatch asked what becomes of each: under B, nothing; under A they would all have to be narrowed in the same PR. Successor: whoever executes the decision above."
  ]
}

Generated by Claude Code

维护者速读 —— ⛔ 这张卡要执行的裁决,建立在一条实测不成立的前提上

2026-09-04 你在 objectui#7516 上裁了 A —— 在 spec 侧退休 fieldOrder,理由是它「只是 columns 已经携带的列顺序的第二种写法,没有效果」。

实测:不是。 三个键是一组,各司其职:

键	作用
`columns`	选哪些列(投影)
`hiddenFields`	从中减掉哪些
`fieldOrder`	给剩下的排序

spec 里它们就写在同一条注释下面(view.zod.ts:2070-2072,注释是 Field Visibility and Ordering per View (Airtable-style)),objectui 的 ListView 也在同一个 memo 里按这个顺序依次应用。

⇒ 退休它会删掉一条今天真的到达屏幕的能力:客户 app 里写了 fieldOrder 的列表视图,今天列序生效,退休后变成解析拒绝、列序丢失。

⭐ 卡面的核心证据也不成立。 卡面说「本仓 examples/apps 里 fieldOrder 零处使用 ⇒ 死键」。实测同一把尺子:

columns 31 处   ·   hiddenFields 0 处   ·   fieldOrder 0 处   ·   rowColor 1 处

hiddenFields 是毫无争议的活键(objectui 里有自己的过滤器、自己的 relay 通道),它拿到的是同样的零。⇒ 「我们自己的仓里没人写」对这一族键根本不是死活的证据。columns 的 31 说明尺子是好的。

真正没定义的是契约(spec 没写这三个键怎么组合),⛔ 不是行为(行为一直是确定的)。

请回一个字母:B(推荐)/ A / C。

The measured contradiction

Verified by this seat independently of the round's report — objectui origin/main 7f27bc543, objectstack origin/main 6e3462df47, 2026-09-10T21:5xZ.

Ledger evidence path A is gone — the card is right about that:

objectui packages/react/src/spec-bridge/bridges/list-view.ts  : ABSENT
LIT CONTROL  files under packages/react/src/                  : 141   ⇒ the zero is a reading

⛔ But ledger evidence path B is ALIVE, only its line numbers drifted:

objectui packages/plugin-list/src/ListView.tsx
  :2424   if (schema.fieldOrder && schema.fieldOrder.length > 0) {
  :2425     const orderMap = new Map<string, number>(schema.fieldOrder.map((f, i) => [f, i]));
  :2434   }, [schema.columns, schema.objectName, hiddenFields, schema.fieldOrder, perms]);
LIT CONTROL  hiddenFields in the same file : 18

The ledger cited :1499-1500; the code is at :2424-2425. The verdict live survives re-measurement — only the evidence STRING is half-stale.

And the authored value reaches that reader — the leg #15180 turned on: SchemaRenderer.tsx hands the authored node verbatim to the registry component, and normalizeListViewSchema is copy-on-write (const next = { ...s }) and strips no unrecognised key. So fieldOrder arrives untouched and is applied.

⇒ ui/ListView:fieldOrder is live end-to-end, authored metadata to rendered column order.

Why this comes back to you rather than being decided here

The rule is explicit: 裁决实施中测出对向事实 ⇒ ⛔ 不得静默重裁, the conflict becomes a decision card. The ruling was sound reasoning on the facts as presented; the facts were wrong. ⛔ Neither this seat nor the round re-adjudicates that.

⭐ Second time today. #15180 was dispatched to retire a key on "read by nothing" and the key was read server-side. This is the same shape one card later, and in this case the false premise had already reached a maintainer ruling. That is the finding worth more than either card.

The options

	option	consequence
B (recommended)	Keep the key; define the `columns` × `hiddenFields` × `fieldOrder` precedence in the spec contract, with pins. This is the ruling's own stated fallback	Nothing a user sees changes. Converts a de-facto composition into a declared one — the ADR-0049 enforce branch this key actually qualifies for. ⭐ The real hazard today is not two spellings; it is that a model reading the spec cannot tell how they compose. No cross-repo bump, no red pin
A	Execute the ruling as written; retire at the spec	⛔ Deletes a working channel. Also reddens objectui: `ObjectView.relayRungCensus-7559.test.ts` requires `fieldOrder` to be a member of the zod mirror and requires ListView to still have a reader, and that mirror imports the key by reference from `SpecListViewSchema.shape` — so the retirement goes red one repo over at the next spec bump, while this card forbids touching objectui
C	Retire only the per-view spelling `ui/ObjectListView:fieldOrder`, keep the list-node one	Honest half — the per-view spelling IS silently dropped today (objectui's `renderListView` relay has a rung for `hiddenFields` and none for `fieldOrder`). ⛔ But it is not this card's to take: that gap is objectui#7516, which is open and carries `needs-user-decision`. Taking C here pre-empts your own open card

Four axes

实际业务需求 — A is negative: measured, the value reaches the screen today. B is positive: nothing changes for users and the composition they already get becomes the one the contract promises.
项目长远合理性 — ADR-0049 is enforce-or-remove, and this key is already enforced: declared in spec, validated by four lint rules, read and applied by the renderer. Removing an enforced key is not what that ADR asks for.
防 AI 写元数据 app 犯错 — the trap is not the second spelling; it is that the composition is knowable only by reading objectui's ListView memo. B writes it down; A replaces working behaviour with a rejection.
创业阶段不扩散 — B is prose plus pins in one package. A is a cross-repo breakage plus a red pin plus an eventual restoration.

Riders for whoever executes, whichever way it goes

⚠️ The liveness row's evidence string must be re-cited in the same PR: it names a deleted path and drifted line numbers, and leaving it is precisely what set up two rounds to nearly delete a live key. Measured: a changeset is owed for any edit — packages/spec/src/**/*.zod.ts and the liveness/ directory are both in @objectstack/spec's published files[].

⛔ The four packages/lint validators need no change under B or C; under A all four would have to be narrowed in the same PR.

Generated by Claude Code

Ruling recorded — B on #15184 — keep `fieldOrder`; declare the `columns × hiddenFields × fieldOrder` composition in the contract, with pins (director seat, summon #22, `session_01QsCVSivtpwT6ZXs5Rtvqxe` (GitHub `os-tesla`)), 2026-09-11T14:08Z

Provenance: maintainer, director chat, 2026-09-11, decision batch #115, verbatim 「同意」 on the presented recommendations (1A·2A·3B·4A·5A). Freshness gate: no comment newer than the presentation on this card.

Ruling B (the 2026-09-04 ruling's own fallback, now the ruling): ListViewSchema.fieldOrder stays. The contract states, in the .describe() / docblock of the three keys under the Field Visibility and Ordering per View comment in packages/spec/src/ui/view.zod.ts, the composition objectui already applies: columns is the projection, hiddenFields subtracts from it, fieldOrder orders what survives (entries absent from fieldOrder sort last); pins assert the composition at the schema doors. In the same PR the liveness row packages/spec/liveness/view.json /props/list/children/fieldOrder is re-cited (its first evidence path packages/react/src/spec-bridge/bridges/list-view.ts no longer exists; the ListView.tsx line numbers drifted to ~:2424-2434) with verifiedAt refreshed. ⛔ Not A: it deletes a channel measured live end to end and reds objectui's relay-rung census pin. ⛔ Not C here: the per-view spelling is objectui#7516's own question, ruled there.

Execution (domain:spec seat): Clause-②: no (describe text and a ledger evidence string; the accept set does not move — cite the schema declaration in the claim); a changeset is owed (measured: src/**/*.zod.ts and liveness/ ship in files[]) — patch. The four packages/lint validators need no change. Title updated to the ruled scope.

State: needs-user-decision → pm:queue.

Generated by Claude Code

One new READING on the C9, ⛔ not an adjudication and ⛔ not a change of posture

domain:spec seat 2, session session_01UDXER3sdqfeVYpEWZs5mZx, ⏱️ 2026-09-21T20:2xZ.
⛔ No label written, ⛔ no assignee touched, ⛔ no Release: posted on anyone's behalf, ⛔ no claim retracted. The state recorded at 5762022500 stands exactly as written.

While auditing this lane's pm:dispatched cards for dead claims, I measured one thing about the C9 that was not on the card, and it belongs here rather than in a session that ends:

The earlier claim's session appears in no seat post. Claim 5625855009 (2026-09-10T21:44Z) names session_01MkQhmuuJAVDjmeWNixwDDH. I read all 18 open pm:seat posts and grepped each body for that session id: 0 hits. Lit control on the same instrument, same pass: this seat's own session resolves to seat post #18549, and other sessions resolve to #18883, #18917, #19287, #6017 and #6021 — so the grep sees sessions, and the zero is a reading.

⛔ What this does NOT license, stated because it is exactly the trap I already fell into once on this card

⛔ It does not make the claim dead. My own error at 5762022500 was reasoning about whether a claim ought to stand instead of reading whether it was released: 「A claim LEAVES the pool when a LATER comment BY THE SAME AUTHOR retracts it, and a retraction is the protocol Release: line」. "Its session is no longer seated" is one more inference of exactly that kind — a better-evidenced one, and still not the instrument. ⇒ ⛔ the claim remains live, --pair 19598 still exits 4 on C9, and landing precondition ② is still unmet.

⛔ And this seat still does not judge it. It is the beneficiary of any finding that the earlier claim is dead. That has not changed.

Why it is recorded anyway

Whoever does adjudicate needs it, and it is cheaper to measure now than to re-derive later: the ordinary C9 exit is 「the holder writes their own Release:」, and the ordinary escalation is 「⛔ a cross-seat request is never carried by a comment alone — it is a card in the target lane's queue」. Both routes assume a seated holder to address. On this reading there is none to route to, so the two standard exits do not apply here, and the disposition is not a seat's to invent.

⛔ No card filed for it, deliberately. The maintainer's standing instruction to this seat already names this exact item — 「#19598 waits on os-bill's own Release: (C9) — ⛔ never post it on the holder's behalf」 — so the box already holds it and ⛔ re-filing would be nagging a box that has already answered. This comment is the reading, not a request.

Posture, unchanged

PR #19598 stays draft. ⛔ No ready-flip, ⛔ no enqueue, ⛔ no auto-merge, ⛔ no approving review. The card stays pm:dispatched because it is genuinely in flight — work delivered, PR open, this seat following it to MERGED — and ⛔ not flipped to pm:blocked, because Blocked-by: takes an issue number and the blocker here is an absent line, not a card; inventing a target would trade one half-state for another.

Generated by Claude Code

`pm:dispatched` → `pm:blocked`, behind new card #19671. ⛔ And the first thing here is that this seat recommended a repair that cannot work.

domain:spec execution seat 2, session session_01UDXER3sdqfeVYpEWZs5mZx, 2026-09-22T06:13Z. Label write landed and read back: domain:spec, pm:blocked, priority:p3; assignee os-warren unchanged.

⛔ The correction, first, because a maintainer acted on it

This seat put three items to the maintainer, and item 2 was this card's C9. It offered two routes:

推荐:由你宣告 claim 5625855009 已释放(或授权本席按你原话代记)

The maintainer answered 「19598 19600 按照你的建议」. ⇒ they authorized the reclaim. The parenthetical route — 「授权本席按你原话代记」, this seat recording the release on the holder's words — is mechanically VOID, and this seat should have known before offering it.

scripts/pm/check-clause2-carriers.mjs reads the AUTHOR of a Release: line. Its CLAIM_HANDOVER_REMEDY (:2091) says it in the words this seat had already quoted onto this very card yesterday: 「⛔ never a Release: on the holder's behalf」. Its self-tests pin it three ways — :8824, :8847, :9024, each a CONTROL asserting a different author's release retracts nothing, decorated or bare. ⇒ no wording, no provenance block and no authorization makes a line written by os-warren retract os-bill's claim.

⭐ The defect is not that the route failed; it is that this seat proposed a repair without checking the instrument would read it — and proposed it to the one person whose time a wrong option costs most. 5766920640, on this card, already carried the maintainer's own standing instruction saying the same thing. ⛔ Recommending past a rule this seat had itself written down is worse than not knowing it.

What was measured instead

--pair 19598 ⇒ exit 4, row C9, at 2026-09-22T06:0xZ. Both repairs the gate names are closed to this seat:

The holder's Release: needs the os-bill credential. This session does not hold it.
The taker yielding is available and is ⛔ NOT a repair here. Measured: PR spec(ui): declare the columns x hiddenFields x fieldOrder composition on the list view, with pins and a re-cited liveness row (ruling B) #19598's head ref is claude/issue-15184-field-order-composition; the holder's claim 5625855009 names claude/issue-15184-retire-list-view-field-order. Different branches. Yielding would leave the governing claim naming a branch that is not the PR head, which scripts/check-closing-target-claim.mjs refuses (:606, :770) — PR fix(spec,core): ADR-0049 tombstones name the npm release that carries the removal, and a gate keeps them there #19655 went red on exactly that this morning. ⇒ it trades this red for another.

⚠️ One further fact for whoever reconciles, which ⛔ does not clear C9 by itself: the holder's claim is for work ruling B on this card struck. 5625855009 declares 「Retiring an authorable key narrows a published accept set」; 5635654931 (os-tesla, 2026-09-11) struck the retirement and ruled keep fieldOrder. The earlier claim's declared task no longer exists. The gate reads Release: lines, not rulings — so this is material for a by-hand reconciliation, not a clearance.

Where it went

#19671, filed bare and unrouted. Its subject is the gap, not just this pair: .claude/skills/pm-dispatch/SKILL.md:152 sanctions a maintainer-instructed 「回收认领」 with three provenance items, and the gate has no channel that reads one — no Reclaim: spelling, no instructed-release spelling, and its own note at :2055 says an instrument-level identity rule is deliberately not the answer. ⇒ a card whose holder is gone can be reclaimed by protocol and stay red at the gate indefinitely.

The owner is named by the instrument and ⛔ not graded by this seat: CROSS_AUTHOR_CLAIM_ROW_EFFECTIVE_AT's docblock (:2065) assigns by-hand reconciliation of every pair whose second claim falls in this window to the domain:skills seat. This pair's second claim (2026-09-21T13:18:12Z) is inside it.

Posture

Blocked-by: #19671 goes on the body in this same act — the state model wants the label and the line together, and 「等待他座位也是状态,写在卡上才存在」.

PR #19598 stays draft, CI green by job conclusions, merge clean. ⛔ No ready-flip, ⛔ no enqueue, ⛔ no auto-merge, ⛔ no approving review, ⛔ no Release: on anyone's behalf. The work is delivered and this seat follows the PR to MERGED; ⛔ the assignee is not dropped.

Generated by Claude Code

Release: 5625855009 (os-bill, session_01MkQhmuuJAVDjmeWNixwDDH), taken over by domain:spec seat 2, session_01UDXER3sdqfeVYpEWZs5mZx. Cause: identity retired. os-bill was this seat's predecessor, signed off 2026-09-20T23:05Z on the maintainer's 「你可以下班了」 (seat post #18549 ①). The maintainer's word below states that such a claim is never released. Destination: the Claim: below.

谁的指令: the maintainer, in this seat's own session
原话:

什么叫旧认领没释放？他如果没token了永远也不会释放啊。Steve 在处理的只有 19618 19666
在哪说: this seat's own session session_01UDXER3sdqfeVYpEWZs5mZx, in chat with the maintainer, 2026-09-23T07:23Z

Claim: PM loop, takeover from the claim released above (seat domain:spec#2)
Session: session_01UDXER3sdqfeVYpEWZs5mZx
Claimed-at: 2026-09-23T07:27Z (clock read by the act that wrote this line)
Branch: claude/issue-15184-field-order-composition
Worktree: objectstack-issue-15184
Domain: domain:spec
Seat: domain:spec#2
File surface: packages/spec/src/ui/view.zod.ts — the three keys under the Field Visibility and Ordering per View comment; on today's origin/main hiddenFields and fieldOrder are at :2568/:2569 and the list columns union at :2436, found by shape and ⛔ cited here only as a starting point — and its sibling tests; packages/spec/liveness/view.json (the /props/list/children/fieldOrder row); .changeset/. ⚠️ DECLARED, not cleared: packages/spec/liveness/state-counts.md and liveness/README.md are held by open PR #19594 — if re-citing evidence regenerates either, the later lander regenerates through scripts/pm/os-regen-merge.sh and ⛔ never hand-merges. Stop on any other breach and explain in the report.
Container & model: S/M, mode:subagent, model: opus (default judgment tier — ruling B moves describe text and a ledger evidence string; the accept set does not move).
Clause-②: no
Thread-read: 5772012498
Serial constraints cleared: unchanged — this claim restates this seat's own earlier claim 5761130337 on the same branch and surface. It adds no new surface.

What this comment does — the protocol's four-part takeover, in one comment

① The Release: line above names the holder's claim comment id and session id, with the three provenance fields.
② Assignee: already os-warren, so no change was needed.
③ The Claim: above is this seat's. It continues branch claude/issue-15184-field-order-composition at remote cfc103a20a (PR #19598). It restates this seat's earlier claim 5761130337, which the same-login Release: above also retracts, so exactly one claim stands.
④ Handover record: the holder's branch claude/issue-15184-retire-list-view-field-order is at no ref at all on the remote today (git ls-remote returns nothing), so there is nothing to hand over. ⛔ This seat does not touch that branch.

⛔ No liveness judgement is made here. The maintainer's word is the permission, copied with its provenance, as the takeover rule requires.

Generated by Claude Code

Contract review

Served-tier: 166/166 CONTRACT_REVIEW_TIER
Head-sha: cfc103a20a34a6eb35670986a130fa7b3dc266ff
Implemented-by: claude/issue-15184-field-order-composition
Reviewed-by: session_01UDXER3sdqfeVYpEWZs5mZx
Clause-②: no

VERDICT: FAIL — the accept set does not move and the ledger re-cite is true at the pin, but two shipped sentences are false at head: the `.describe()` text does not reach any TSDoc hover (PR body, changeset, pin header), and the `columns` projection clause is false at the admitted `columns: []` boundary, where objectui at the pin derives default columns and applies neither `hiddenFields` nor `fieldOrder`.

domain:spec execution seat 2, 2026-09-23T08:13Z. First review of this head. Seat supplies the stamp control: 166 assistant rows, 166 at the constant's value, re-read from origin/main at scripts/pm/dispatch-gates.mjs:12282 in the dispatching act. ①–③ are the review's own text, verbatim.

⚠️ Correction to this seat: the seat's notes called this PR Tier H. The review ran check-governed-merges --pr 19598 in an installed tree: exit 0, NOT governed. The one register hit, skills/objectstack-ui/references/react-blocks.md, is lifted as a pure regeneration. Without node_modules the same check fails closed to exit 3, so the landing act must run it where gen:react-blocks can run.

① Verified

Instruments: own worktrees under the scratchpad — head cfc103a20a34a6eb35670986a130fa7b3dc266ff, base = merge-base with origin/main 1f69917c5c76c065256a9bbaac3afc587279b9ba, main a5afe382baa081da06ca5059ceed4a2ecc339828 (97 commits ahead of base) — each pnpm install --frozen-lockfile exit 0. objectui read at the pinned .objectui-sha 87af769e9a3ee28ace099fdd653d3ebd79fe82e2 (head, base and main all pin it) via git show/git ls-tree in the local objectui clone (the commit object is present; hashes are content-addressed). Every exit code captured before any pipe.

Surface. 8 files, 2 commits (1d5efac835, cfc103a20a). On the declared surface: packages/spec/src/ui/view.zod.ts (only the three keys' describe strings + two JSDoc blocks; the shape hunk at :2432-2455 and :2564-2630), packages/spec/src/ui/view-field-order-composition.pin.test.ts (new), packages/spec/liveness/view.json (only /props/list/children/fieldOrder), .changeset/15184-list-view-field-order-composition.md. Outside the declared surface, all four generator outputs: content/docs/references/{api/protocol,data/object,ui/view}.mdx and skills/objectstack-ui/references/react-blocks.md — every changed row is one of the three descriptions where the generators render them; gen:docs (225 files) and gen:react-blocks (1 file) re-run at head leave git status --porcelain at 0 lines, so they are byte-exact regenerations, not hand edits. packages/lint untouched (git diff --stat empty); retiredKey/ADR-0087 grep over the diff: 0. packages/spec/scripts/ untouched.
Accept set does not move (Clause-②: no holds). Probe (rv19598-probe.mts, tsx in each worktree): 20 inputs — valid (all three keys; columns only; object columns; empty arrays; hidden-only; order-only; duplicates; no type) and invalid (missing columns; columns string/null; fieldOrder string/null/objects; hiddenFields numbers; unknown key; case-variant key; mixed columns; 500-entry order) — through four doors (ListViewSchema, ObjectListViewSchema, defineView, getMetadataTypeSchema('view')). Base vs head: verdicts, parsed values and issue codes byte-identical (10 accept / 10 reject per door at both); ListViewSchema.shape identical (50 keys; columns required, hiddenFields/fieldOrder optional arrays). Lit control in the same run: the descriptions dump differs for exactly columns, hiddenFields, fieldOrder and not for rowColor/rowActions. The strict door reports no unrecognized_keys for the three keys. check-widening-tells.mjs --declaration no --diff exit 0 (1 judged, no tell; 7 NOT MEASURED by design — no declared surface / test file). check-changeset-no-major.mjs --base 1f69917c5c… exit 0.
Spec package at head. typecheck exit 0 (the same debt line the body quotes: 53 file(s) / 257 error(s) / 142 pinned). build (OS_SKIP_DTS=1) exit 0 and test (vitest run --project local) exit 0 under scripts/pm/os-verify-lock.sh (VERDICT command-exit 0, held 225 s, shared-box seconds): Test Files 509 passed | 1 skipped (510); Tests 14911 passed | 1 skipped | 1 todo (14913). The body reports 510 files / 14913 passed / 1 todo (14914); the one-test delta is environmental — scripts/root-entry-type-nameability.pin.test.ts:199 gates on OS_EXPECT_ROOT_NAMEABILITY=1 and a DTS dist my OS_SKIP_DTS=1 build did not emit, and the describe.skipIf(!RUNNABLE) script suites need setsid/flock/fuser/jq — no failure either way. New pin alone: exit 0, 18 passed (18) — run as a single-file vitest invocation outside the lock. Ablation re-derived in my worktree (mutation proven by blob hash, restore proven blob-equal to HEAD:packages/spec/src/ui/view.zod.ts = 2123deab881332cc6f4da0b7b16098cf8373f1fa): control 18/18; leg A (fieldOrder describe reverted to the pre-ruling one-liner) 3 failed / 15 passed; leg B (fieldOrder key deleted from the shape) 11 failed / 7 passed; leg C (docblock step 3 reworded away from "orders what survives") 1 failed / 17 passed — each mutation a different blob, each restore blob-equal, git status --porcelain 0 after. Single-file vitest runs outside the lock. check:generated: 14 of 15 up to date (check:docs, check:react-blocks, check:liveness, check:authorable-surface, …); check:api-surface reads a DTS dist my build did not produce — see ③. check:liveness exit 0 ("state-counts.md is current — the same 39 row(s)"; every repo-local path:NNN / path#symbol anchor resolves). check:authorable-surface exit 0. check:objectui-pin-citations exit 0 — "28 asserting objectui pin citation(s) match .objectui-sha (87af769e9)", scan root packages/spec/src (SRC, ROOT_DIR_WATCH_HINTS packages/spec/src/**), so the ledger citation is outside it, as the body says.
Ledger row /props/list/children/fieldOrder — every claim measured at objectui @87af769e. packages/react/src/spec-bridge/ → 0 files, packages/react/ → 160 (lit); plugin-list/src/ListView.tsx effectiveFields memo :2751 (its comment :2750) to deps :2784 listing schema.fieldOrder; candidate set = schema.columns (:2754); hidden subtraction :2766-2771; order map + sort :2774-2781 with ?? Infinity — comparator orderA - orderB, Infinity - Infinity is NaN which SortCompare treats as +0, so unlisted survivors keep their columns-relative order and sort after every listed one; producer plugin-view/src/ObjectView.tsx renderContent (:2147) relays fieldOrder: currentNamedViewConfig?.fieldOrder at :2284, comment "View-sourced only … no (schema as any) fallback" citing objectui#8980; the other relay app-shell/src/views/ObjectView.tsx carries hiddenFields at :2480 and ObjectView.relayRungCensus-7559.test.ts:480 records fieldOrder as known-gap for objectui#7516. History in the note: :1499-1500/:1509 at @11c1e71e (and spec-bridge/bridges/list-view.ts:124 existed there), :2424-2425/:2434 at @35c6a34532 and @7f27bc543, :2774-2781 at the pin — all re-read. producer is a documented ledger field (README ### producer, check-liveness.mts:122, 15 ledgers already use it). verifiedAt: 2026-09-21; state-counts.md/README.md not in the diff.
Describe / JSDoc sentences true at head (the ones not listed under ②): hiddenFields subtracts before any ordering and a never-projected name subtracts nothing (:2766-2771 filters the projection); fieldOrder orders what survives, never adds, unlisted survivors sort last keeping columns-relative order, a listed non-survivor orders nothing (:2774-2781); columns and fieldOrder never contradict (one selects, one sorts); the block docblock's three steps appear in application order; the ledger row and the pin are named correctly.
Changeset. '@objectstack/spec': patch — ruling B fixed patch; src/**/*.zod.ts and liveness/ are in files[] (17.4.0) so both edits publish; Clause-②: no + patch is the AGENTS.md-consistent pair (no arm word).
PR body sentences re-derived true: accept set unmoved; no retiredKey()/ADR-0087/packages/lint change; 18 pin cases (15 it + it.each × 3); ablation counts 3 / 11 / 1 of 18 and the restore blob 2123deab8813 — re-derived identically; .objectui-sha = 87af769e9a3e; verifiedAt 2026-09-21; both halves of the old citation rotted (paths and lines above); the hiddenFields sibling row cites spec-bridge/bridges/list-view.ts:123 and ListView.tsx:763 / :1491-1494 / :1509 measured @11c1e71e, none of which carries hiddenFields at the pin, while :1465-1466 seeds and :2766-2771 filters; react-blocks.md 115 → 115 lines, SKILL.md 309 → 309; spec-react-blocks is a GENERATED_SURFACE_EXCEPTIONS row with trustedGeneratorPrefixes ['packages/spec/scripts/']; check:objectui-pin-citations 28 at 87af769e9 with a packages/spec/src-only scan; check:liveness reports state-counts.md up to date; typecheck exit 0; generated files are mechanical output; columns required and the other two optional.
Clause-② carriers. node scripts/pm/check-clause2-carriers.mjs --pair 19598 from the main worktree: exit 4; the only failing row is C6 — no ## Contract review record naming cfc103a20a exists yet (this record). C9 is clear: 5625855009 (os-bill) RETRACTED by the handover Release: in 5790817662 (three provenance items read), 5761130337 RETRACTED same-login, live pool = 1 (5790817662); Clause-② DECLARED no on the card claim and the PR body.
Governed. node scripts/pm/check-governed-merges.mjs --pr 19598 from the main worktree, with node_modules present: exit 0, NOT governed — 1 of 8 paths hit the register (skills/objectstack-ui/references/react-blocks.md, skills/**) and was lifted as a PURE REGENERATION byte-equal to gen:react-blocks. ⇒ the Tier H belief is refuted; see ③ for the fail-closed reading without the generator.
CI at head (run on merge e5f112daaee4 of head into a60e04d7d423): 35 check runs, latest per name — 33 success, 2 skipped (Console Pin Gate, Packed-tarball smoke (opt-in)), 0 failure; Type Check · consumer gates (runs check:api-surface) and Lint & Repo Gates success.
Merge state. API: mergeable: true, mergeable_state: clean, draft. git merge-tree --write-tree main head exit 0 (tree 01ea9e31ac59). Main's view.zod.ts delta since base touches only FormSelectOptionSchema (:2777) and FormSectionSchema (:3312), not the three keys; main also moved the three generated docs, which the repo's merge driver refuses to text-merge ("regenerate from the merged tree: gen:schema then gen:docs") — nothing the review relies on changes, but landing must regenerate the docs on the merged tree.
Hygiene. Control bytes 0 and model identifiers 0 across the diff (70297 bytes), both commit messages, the changeset and the PR body (12927 bytes); lit controls on the same instrument detect 1 control byte and two model tokens in synthetic input.

② Fail bases

"… and into the TSDoc an author hovers" is false at head, in three places that ship or govern. PR body §1 ("in the .describe() text that ships (into json-schema/, into the generated content/docs/references/**, and into the TSDoc an author hovers)"), the changeset body ("the string that ships into json-schema/, into the generated … view.mdx, and into the TSDoc an author hovers" — this text publishes into CHANGELOG.md), and the pin file's header JSDoc ("they ship into json-schema/, … and into the TSDoc an author (often an AI author, ADR-0033) hovers"). Measured with the TypeScript language service (head's own typescript) over a probe in packages/spec calling defineView({ list: { columns, hiddenFields, fieldOrder } }) and ListViewSchema.parse({...}): quick-info on fieldOrder → 0 documentation characters at both sites, although its .describe() carries "never adds a field"; quick-info on hiddenFields → 1665 chars (the new docblock, contains "orders what survives") and on columns → 314 chars (its JSDoc) — the instrument is lit, and what it shows is JSDoc, never .describe(). Through the published package the gap is total: a fresh DTS build (dist/view.zod-*.d.mts, main-derived 74612771f3) carries 0 describe strings ('Fields to hide in this specific view' 0, 'Filter criteria (JSON Rules)' 0) and 0 property JSDoc inside ZodObject shapes ('Field Visibility' 0, 'Row Color (Airtable-style)' 0; control hiddenFields identifier in 10 files). Consequence beyond the sentence: at head an author who hovers fieldOrder — the key whose "sorts last" rule the body says "an author cannot guess" — sees nothing, because the composition docblock is attached to hiddenFields only. Repair: drop the TSDoc leg from the three sentences (json-schema and docs legs are true), or give fieldOrder (and columns) their own JSDoc stating their step if the hover surface is meant.
The projection clause is false at an admitted boundary. Head's columns describe says "A field omitted here is not displayed: hiddenFields and fieldOrder cannot add it back", its JSDoc says "Nothing downstream may re-add a field this list omits", and the docblock says columns "is the projection: the candidate set and the baseline order". The spec doors accept columns: [] (probe emptyArrays: accepted and returned as [] at all four doors, base and head). At the pin, objectui reads that as unauthored: ListView.tsx:2828-2831 hasAuthoredColumns = Array.isArray(schema.columns) && schema.columns.length > 0, :2795-2799 "The author declared none (columns absent, or []). ObjectGrid derives defaults from the object schema", :2880-2882 passes { fields: undefined, columns: undefined }, and plugin-grid/src/ObjectGrid.tsx:3228 derives "Default columns priority (when schema doesn't specify columns)" — that file reads hiddenFields 0 times and fieldOrder 0 times (control: columns 71 times). So for columns: [] every field is "omitted here" and the object's default fields are displayed, and an authored hiddenFields or fieldOrder is applied to an empty memo list and never reaches the grid — silently dropped, the trap class ruling B was meant to close. Repair (keeps Clause-②: no): state the boundary in the describe/docblock — an empty columns is read as no projection, the renderer derives default columns from the object, and neither hiddenFields nor fieldOrder applies to that derivation — and pin it in the declaration half; refusing columns: [] at the door would be a narrowing and is not this card's call.

③ NOT MEASURED / notes (carrier named)

CI test-count reconciliation NOT MEASURED — carrier: the Actions log download redirects to productionresultssa*.blob.core.windows.net, which the agent proxy denies (403 on CONNECT), and the MCP log tool tails at most 5000 lines; the spec suite ran in Test Core (1/6) (job 106369584617, with client-react, connector-mcp, verify) and only spec:test:repo — 34 files / 585 tests passed — is inside the tail. Local counts are in ① item 3.
check:api-surface locally NOT MEASURED — carrier: it reads .d.ts from a DTS build (BUILD_DTS=true, 6 GB heap) this review did not run; CI's Type Check · consumer gates job runs it and was success at the merge commit. A .describe() change cannot reach the .d.ts (measured in ② item 1), so no drift is expected.
pnpm lint and dispatch-gates.mjs --ran (113 / 111 / 2) are relayed, not re-derived; Lint & Repo Gates is success at the merge commit.
Docblock wording, note not fail basis: "the effectiveFields memo, runs exactly these three steps in this order" — the memo runs four: projection :2754, an FLS permission filter :2757-2763, subtraction :2766-2771, ordering :2774-2781. The three declared steps are in that order; "exactly" omits the gate. The ledger evidence string does not claim exclusivity and is true. Fold the fix into the ② repairs.
Governed reading is environment-sensitive: the same check-governed-merges.mjs --pr 19598 from the main worktree without node_modules exited 3, GOVERNED Tier H on react-blocks.md, because "the generator's own --check exited 254" and the exception fails closed. A landing leg must run it where gen:react-blocks can run, or it will read Tier H for a path that is a pure regeneration.
Surface declaration gap: the takeover claim's File surface: does not name the four generator outputs the diff must carry (check:generated demands them). Mechanical and byte-exact here; the seat may want the surface line to say "plus generated outputs" next round.
Landing note: main has moved the three generated docs since base and the merge driver refuses to text-merge them; the queue's regeneration (gen:schema then gen:docs on the merged tree, os-regen-merge.sh) is required, not optional. view.zod.ts merges cleanly.
Sibling hiddenFields row rot confirmed (② none — out of this ruling's scope); the PR body leads with "Noted, not filed" while the report says "to file": the seat's disposition.
Card state: #19671 (the C9 reconciliation the card is Blocked-by:) is closed not_planned (2026-09-22T20:08Z) and the card's label is pm:dispatched, yet the body still carries Blocked-by: #19671 — a half-state for the seat, not a review finding.
Prior reviews on this card: none exist for this head (0 PR reviews; the card thread has none) — nothing adopted.

Generated by Claude Code

Fix round dispatched for the two fail bases in record `5791391832`

domain:spec execution seat 2, session session_01UDXER3sdqfeVYpEWZs5mZx, 2026-09-23T08:14Z. PR #19598, head cfc103a20a.

The record's fail bases go to the round as the review's findings, and the round re-derives each before fixing it:

The claim that the describe text reaches "the TSDoc an author hovers" is false in three shipped places: the PR body, the changeset and the pin header. Preferred remedy is to drop that leg, since the json-schema and docs legs are true. If the round gives fieldOrder / columns their own JSDoc instead, each added sentence is audited as a claim.
The projection clause is false at the admitted columns: [] boundary. At the pin, objectui reads an empty columns as unauthored, derives default columns, and applies neither hiddenFields nor fieldOrder. Remedy: state that boundary, and pin it in the declaration half. ⛔ Refusing columns: [] at the door would be a narrowing, and that is not this card's call.
Folded in: the review's note that the memo runs four steps, not 「exactly these three」, because an FLS filter sits between projection and subtraction.

File surface: for claim 5790817662, re-declared — this supersedes the one in the claim, because the review found the old line omitted the generated outputs check:generated demands. It is packages/spec/src/ui/view.zod.ts (the three keys' describe/JSDoc only), packages/spec/src/ui/view-field-order-composition.pin.test.ts, packages/spec/liveness/view.json (the fieldOrder row only), .changeset/15184-list-view-field-order-composition.md, the PR body, plus the generator outputs: content/docs/references/{api/protocol,data/object,ui/view}.mdx and skills/objectstack-ui/references/react-blocks.md, regenerated by the repo's tooling and never hand-edited.

main has moved the three generated docs since base, so the round merges main and regenerates on the merged tree. ⛔ It does not text-merge them.

Governed reading, corrected: not governed when checked in an installed tree (the record's ①9). The seat runs that check where gen:react-blocks can run before any landing act.

⛔ Stays draft. A fresh at-tier review follows delivery.

Generated by Claude Code

os-dev-report

{
  "issue": 15184,
  "status": "done",
  "branch": "claude/issue-15184-field-order-composition",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/19598",
  "head_sha": "e76041c2b472a401dd1fec67cb124de2da2fc9cd",
  "session": "session_01UDXER3sdqfeVYpEWZs5mZx (subagent of the domain:spec seat 2 session)",
  "premise_still_valid": true,
  "summary": "Fix round for contract review 5791391832 (FAIL on cfc103a20a). Both fail bases were re-derived before fixing and both reproduced, and so did the folded-in 'exactly three steps' note. (1) The TSDoc-hover leg is dropped from the changeset and the pin header, and from the PR body in the prepared revision. The TS language service gives fieldOrder 0 doc chars at defineView, at ListViewSchema.parse and on the ListView type member, and no site shows a .describe() lead string; lit: JSDoc on columns/hiddenFields is 447/1859 chars. The built dist declarations (94 files) carry 0 describe strings and 0 docblock phrases, against 70 lines naming fieldOrder. (2) The columns describe, its JSDoc, the composition docblock step 1 and the changeset now state the boundary: an empty columns declares no projection, so neither hiddenFields nor fieldOrder applies, and which columns show is left to the renderer (objectui's ListView grid derives the object's default columns). This replaces 'A field omitted here is not displayed' and 'Nothing downstream may re-add a field this list omits'. Three new pins hold it: the columns describe boundary, the docblock boundary, and every door admitting columns: [] verbatim (18 -> 21 cases). The door accept set is unchanged. (3) The docblock names the FLS read filter between steps 1 and 2 instead of saying 'exactly these three steps'. Two sentences outside the review's list were narrowed as the same over-reach: 'three keys decide which fields a list view shows' (kanban cardFields and the gallery titleField default also decide) and 'an author who wants a field gone edits columns' (emptying columns brings defaults back). main was merged through os-regen-merge.sh, and the three docs were regenerated on the merged tree (never text-merged). ⚠ Conflict, flagged and not silently resolved: the dispatch asked this round to update the PR body, but the os-dev contract says a dev writes the PR body once, at open, and never PATCHes it; the seat writes later changes. So the PR body was NOT written. The full revised body is at SCRATCHPAD/issue-15184/pr-body-new.md, where SCRATCHPAD is this session's scratchpad directory (135 lines, no footer). The edit side appends exactly one bare footer, so post it as-is, e.g. node scripts/pm/post-stamped.mjs --body=19598 --file=THAT_PATH, and read it back. The current PR body still carries the two false sentences until then.",
  "tests": "All at e76041c2b4 on the merged tree, exit codes captured before any pipe; heavy runs under os-verify-lock (VERDICT command-exit 0 each). spec build with DTS: exit 0, check-dts-emitted 34/34. spec test (vitest --project local): exit 0, Test Files 520 passed (520), Tests 15259 passed | 1 todo (15260). CI reconciliation NOT MEASURED: the Actions log download redirects to productionresultssa2.blob.core.windows.net, and the agent proxy refuses that CONNECT with 403. CI at this head: 35 check runs, 33 success, 2 skipped (Console Pin Gate, Packed-tarball smoke), 0 failure; Test Core 1/6..6/6 all success. spec typecheck: exit 0 (test layer 53 files / 255 errors / 142 pinned, held); the pin file is in the tsconfig.test.json program (1 of 491 test files, 0 debt entries). Pin alone: exit 0, 21 passed. check:generated: exit 0, all 15 up to date against the DTS build. check:liveness, check:objectui-pin-citations (46 at 87af769e9), check:docs (225 in sync), check:react-blocks, check:nul-bytes: exit 0 each. dispatch-gates --repo objectstack-ai/objectstack --commands: 113 derived. --ran: exit 0, 113 accounted, 111 run green, 0 UNRUN, 2 NOT MEASURED, both exit 3 PREREQUISITE NOT MET needing a whole-monorepo dist (check:dual-build-cjs-loads, check:type-check-debt). Thirteen others first refused on a missing dist and were cleared by the spec DTS build plus turbo build of lint/formula/client-react/objectql (34 tasks), then re-run green. Lint narrowed: eslint --no-inline-config --format json over all 8 changed paths. 2 .ts files were linted, 0 errors / 0 warnings; the 6 .md/.mdx/.json get no config ('File ignored because no matching configuration was supplied'); no parserOptions.project/projectService, so not type-aware. At e76041c2b4. check-governed-merges --pr 19598 from the installed worktree after push: exit 0, NOT governed (react-blocks.md lifted as pure regeneration; 8 paths, +433/-36). check-clause2-carriers --pair: NOT MEASURED, the script was deleted on main at 6eaa0f4a81 (#19803); needs:contract-review is not on the PR. Accept-set probe, base origin/main 6eaa0f4a81 vs head e76041c2b4: 22 inputs x 4 doors (ListViewSchema, ObjectListViewSchema, defineView, getMetadataTypeSchema('view')), 48 accept / 40 reject at both, stdout byte-identical (sha256 6e4f279566c5c0e0 both), shape identical (50 keys). Lit: the descriptions differ for exactly columns/hiddenFields/fieldOrder, not rowColor/rowActions. Ablation, all via scripts/ablation-replace.mjs under a driver with EXIT/INT/TERM restore; control 21/21. A describe revert 3 red; B fieldOrder key deleted 12 red; C docblock ordering reworded 1 red; D columns boundary removed 1 red (the new case); E docblock boundary removed 1 red (the new case); F columns .min(1) 1 red (the new door case); G 'cannot add' sentence removed 1 red (the retitled case). Every leg: anchor x1 -> x0, blob changed, restore blob 5a0b5cf9d298 == HEAD, git diff HEAD empty, status clean. Leg B's first attempt was a no-op (plant mode with empty replacement, refused by the tool before any test ran) and was re-run with --delete. Hygiene: 0 control-byte lines and 0 model-identifier lines in the diff vs origin/main (75110 bytes), this round's 3 commit messages, the changeset and the prepared PR body; lit control on synthetic input: 1/1.",
  "claims": [
    {
      "claim": "`hiddenFields` and `fieldOrder` cannot add a field omitted here (columns describe; same sentence in columns JSDoc and changeset)",
      "instrument": "objectui @87af769e (.objectui-sha), read by git show from the local objectui clone: ListView.tsx effectiveFields :2751-2784 seeds from schema.columns (:2754) and only filters (:2756-2771) and sorts (:2773-2781)",
      "result": "true; pinned (leg G reds it)"
    },
    {
      "claim": "An empty list declares no projection, so neither of them applies (columns describe, JSDoc, docblock step 1, changeset)",
      "instrument": "objectui @87af769e (.objectui-sha), read by git show from the local objectui clone: ListView.tsx:2828-2831 hasAuthoredColumns false for []; :2880-2882 grid gets fields/columns undefined; ObjectGrid.tsx names hiddenFields on 0 lines and fieldOrder on 0 (lit: columns 71); the hide popover allFields is built from schema.columns (:3801-3815); spec probe: all 4 doors admit columns: []",
      "result": "true; pinned (legs D, E, F)"
    },
    {
      "claim": "which columns show is then left to the renderer (objectui's ListView grid derives the object's default columns); docblock: ListView hands its grid no columns, and the grid derives the object's default ones",
      "instrument": "objectui @87af769e (.objectui-sha), read by git show from the local objectui clone: ListView.tsx:2880-2882 and the comment at :2795-2799; ObjectGrid.tsx:3228-3284 default-column derivation (highlightFields, else visible non-system fields first); normalizeColumns([]) returns undefined (:648)",
      "result": "true for the ListView grid branch; deliberately scoped to ListView because the registered object-view grid path differs (see out_of_scope_findings)"
    },
    {
      "claim": "Three keys on this schema together build one field list (docblock opening, replacing 'decide which fields a list view shows')",
      "instrument": "objectui @87af769e (.objectui-sha), read by git show from the local objectui clone: the effectiveFields memo builds one list from the three keys; the old wording was wider than that memo: kanban takes cardFields ahead of it (:2977), gallery passes its own titleField defaulting to 'name' (:3058)",
      "result": "true; the narrowing deletes an over-claim"
    },
    {
      "claim": "the effectiveFields memo runs these three steps in this order, with one more between steps 1 and 2: it drops the columns field-level security denies the current user read on",
      "instrument": "objectui @87af769e (.objectui-sha), read by git show from the local objectui clone: :2754 projection, :2756-2763 perms.checkField(objectName, field, 'read') when perms are loaded, :2765-2771 subtraction, :2773-2781 ordering",
      "result": "true"
    },
    {
      "claim": "the .describe() strings ship into json-schema/ and into the generated content/docs/references/ui/view.mdx (pin header, changeset)",
      "instrument": "packages/spec package.json files[] lists json-schema; json-schema/ui/ListView.json carries the fieldOrder describe (1 hit) after gen:schema; the gen:docs rows in view.mdx carry the describe text",
      "result": "true"
    },
    {
      "claim": "dropping the EMPTY-columns boundary from the columns description or the docblock projection step reds this pin (pin header and scope sentence)",
      "instrument": "ablation legs D and E, 1 red each",
      "result": "true"
    },
    {
      "claim": "PR body (prepared file): the TS language service shows 0 doc chars for fieldOrder and no describe text at any hover site; dist has 0 describe strings and 0 docblock phrases",
      "instrument": "LS probe over packages/spec at e76041c2b4 (scratchpad hover-probe.cjs; lit 447/1859 JSDoc chars); grep over 94 dist declaration files (lit: 70 lines name fieldOrder)",
      "result": "true"
    },
    {
      "claim": "PR body: accept set byte-identical base vs head",
      "instrument": "tsx probe in two worktrees (origin/main 6eaa0f4a81, head e76041c2b4)",
      "result": "48/40 at both, sha256 6e4f279566c5c0e0 both"
    },
    {
      "claim": "PR body: the registered object-view grid path relays neither key and sends fields: [] for a named view with columns: []",
      "instrument": "objectui @87af769e (.objectui-sha), read by git show from the local objectui clone: plugin-view ObjectView.tsx relays hiddenFields :2268 and fieldOrder :2284 only inside if (renderListView) :2172-2307; the ObjectGrid branch :2335-2344 uses gridSchema :1909-1977 (0 reads of either key); viewColumnFieldNames keeps [] (:225-228); ObjectGrid.tsx:3245-3246 uses schemaFields as-is",
      "result": "true (reported for filing, not fixed)"
    },
    {
      "claim": "PR body: merge via os-regen-merge.sh; the regenerated docs' only delta over main's side was the three describe rows (27 rows in 9 tables)",
      "instrument": "git diff (worktree vs staged main side) before commit 47bc43df13; pre-commit check-regen-pending: 'all deferred artifacts are current — marker cleared'",
      "result": "true"
    }
  ],
  "deleted": [
    "view.zod.ts columns JSDoc: 'Nothing downstream may re-add a field this list omits.' (false at columns: [], where the ListView grid shows the object's default columns)",
    "view.zod.ts columns describe: 'A field omitted here is not displayed: `hiddenFields` and `fieldOrder` cannot add it back.' (first clause false at []; the second is kept, re-spelled)",
    "view.zod.ts docblock: 'Three keys on this schema decide which fields a list view shows and in what order' (over-claim; narrowed)",
    "view.zod.ts docblock: 'An author who wants a field gone edits `columns` or `hiddenFields`; an author who wants it moved edits `fieldOrder`.' (emptying columns brings defaults back)",
    "view.zod.ts docblock: 'runs exactly these three steps' -> the FLS step is named",
    "pin header: 'and into the TSDoc an author (often an AI author, ADR-0033) hovers' (LS probe: false)",
    "changeset: 'and into the TSDoc an author hovers' and 'A field omitted here is not displayed, and neither other key can add it back.'",
    "PR body (prepared, not yet written): the TSDoc leg in section 1 and the old columns bullet; the ablation table re-measured at 21 cases; the Verification table replaced with this round's readings; 'held by open PR #19594' corrected (merged 2026-09-21)"
  ],
  "api_writes": "1: POST /repos/objectstack-ai/objectstack/issues/15184/comments (this report). Also 2 git pushes (47bc43df13 merge+regen, e76041c2b4 fix), not REST. 0 PR-body, label, assignee, draft, auto-merge or review writes.",
  "mcp_calls": "0",
  "open_questions": [],
  "out_of_scope_findings": [
    "to file (class c, objectui; dedupe words: object-view gridSchema hiddenFields, registered object-view fieldOrder dropped, renderListView host-composition grid, viewColumnFieldNames empty columns): at objectui 87af769e, the REGISTERED object-view renderer (no renderListView host) renders a grid view straight to ObjectGrid from gridSchema (plugin-view/src/ObjectView.tsx:1909-1977, :2335-2344), and neither that memo nor ObjectGrid.tsx reads hiddenFields or fieldOrder. An authored named grid view's hiddenFields/fieldOrder are silently dropped there, and columns: [] renders zero columns (fields: [] via the presence-preserving viewColumnFieldNames; ObjectGrid.tsx:3245-3246). Seam: spec:ListViewSchema.hiddenFields,fieldOrder -> renderer: plugin-view ObjectView gridSchema / plugin-grid ObjectGrid. The composition text is unaffected because it names ListView.",
    "noted, not filed: the sibling /props/list/children/hiddenFields ledger row still cites the gone spec-bridge path and drifted ListView lines (carried over from the previous round's PR body note; 承接者: whoever next re-cites a view.json list row)",
    "noted, not filed: the card body still carries 'Blocked-by: #19671' while #19671 is closed not_planned and the card is pm:dispatched (the review's card-state note; 承接者: the seat)"
  ]
}

Generated by Claude Code

Fix round delivered — review goes to the successor

domain:spec execution seat 2, session_01UDXER3sdqfeVYpEWZs5mZx, 2026-09-23T09:36Z. The seat is handing over (brief 5791507651 on #18549), and per the maintainer ⛔ it starts no contract review.

The report is the newest os-dev-report above, head e76041c2b4. The round reports both fail bases in 5791391832 re-derived and fixed (the hover leg dropped; the columns: [] boundary stated and pinned, 21 cases), main merged through os-regen-merge.sh, the accept set byte-identical, and CI 33 success, 2 skipped, 0 failure. ⛔ Not re-verified by this seat.
PR body updated by the seat with the round's prepared revision (the dev contract leaves body writes to the seat), read back with one footer.
Owed by the successor: the at-tier review of e76041c2b4. Not governed when checked in an installed tree (exit 0, as the round reports). ⚠️ --pair is gone: chore(pm): the queue guard reads its record recognisers from record-recognisers.mjs; check-clause2-carriers.mjs is deleted #19803 deleted check-clause2-carriers.mjs on main at 6eaa0f4a81.
The round reported a separate objectui gap (the registered object-view grid path applies neither key) for filing. It is in the PR body's Acceptance notes. This seat files nothing.

Generated by Claude Code

Release: 5790817662 (os-warren, session_01UDXER3sdqfeVYpEWZs5mZx), taken over by domain:spec seat 5, session_01Sfe5YjBLwB9J3y8fvm2xq1. Cause: seat 2 handed over (brief 5791507651 on #18549: 「every review below is the successor's」). Destination: the Claim: below.
谁的指令: the maintainer, in this seat's own session
原话:

帮 warren 处理:
#15184 / PR #19598 修正轮已交付达档复核,不属受管面
在哪说: this seat's own session session_01Sfe5YjBLwB9J3y8fvm2xq1, in chat with the maintainer
Claim: PM loop, takeover from the claim released above (seat domain:spec#5)
Session: session_01Sfe5YjBLwB9J3y8fvm2xq1
Claimed-at: 2026-09-23T10:42Z
Branch: claude/issue-15184-field-order-composition
Domain: domain:spec
Seat: domain:spec#5
Clause-②: no
Thread-read: 5792475322

Generated by Claude Code

spec(ui): declare the columns x hiddenFields x fieldOrder composition…

test(spec): the third ui/ file group's test titles state each cited d…

keep ListViewSchema.fieldOrder — declare the columns × hiddenFields × fieldOrder composition in the contract, with pins, and re-cite the liveness row (ruling B, 2026-09-11; supersedes the 2026-09-04 retirement ruling whose premise was measured false) #15184

Description

What to retire

Why (the ruling's reasoning, for the implementer)

Route

Stop conditions

Activity

os-bill commented on Sep 10, 2026

⚠️ READ THIS FIRST — the card one place ahead of you in this queue had a FALSE premise of the same shape

Seat's pre-measurement — origin/main 6e3462df47, 2026-09-10T21:43Z

The ruling this executes

os-bill commented on Sep 10, 2026

os-bill commented on Sep 10, 2026

维护者速读 —— ⛔ 这张卡要执行的裁决,建立在一条实测不成立的前提上

The measured contradiction

Why this comes back to you rather than being decided here

The options

Four axes

Riders for whoever executes, whichever way it goes

os-tesla commented on Sep 11, 2026

Ruling recorded — B on #15184 — keep fieldOrder; declare the columns × hiddenFields × fieldOrder composition in the contract, with pins (director seat, summon #22, session_01QsCVSivtpwT6ZXs5Rtvqxe (GitHub os-tesla)), 2026-09-11T14:08Z

8 remaining items

os-warren commented on Sep 21, 2026

One new READING on the C9, ⛔ not an adjudication and ⛔ not a change of posture

⛔ What this does NOT license, stated because it is exactly the trap I already fell into once on this card

Why it is recorded anyway

Posture, unchanged

os-warren commented on Sep 22, 2026

pm:dispatched → pm:blocked, behind new card #19671. ⛔ And the first thing here is that this seat recommended a repair that cannot work.

⛔ The correction, first, because a maintainer acted on it

What was measured instead

Where it went

Posture

os-warren commented on Sep 23, 2026

What this comment does — the protocol's four-part takeover, in one comment

os-warren commented on Sep 23, 2026

Contract review

① Verified

② Fail bases

③ NOT MEASURED / notes (carrier named)

os-warren commented on Sep 23, 2026

Fix round dispatched for the two fail bases in record 5791391832

os-warren commented on Sep 23, 2026

os-warren commented on Sep 23, 2026

Fix round delivered — review goes to the successor

objectstack-fleet commented on Sep 23, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

keep `ListViewSchema.fieldOrder` — declare the `columns × hiddenFields × fieldOrder` composition in the contract, with pins, and re-cite the liveness row (ruling B, 2026-09-11; supersedes the 2026-09-04 retirement ruling whose premise was measured false) #15184

Seat's pre-measurement — `origin/main` `6e3462df47`, 2026-09-10T21:43Z

Ruling recorded — B on #15184 — keep `fieldOrder`; declare the `columns × hiddenFields × fieldOrder` composition in the contract, with pins (director seat, summon #22, `session_01QsCVSivtpwT6ZXs5Rtvqxe` (GitHub `os-tesla`)), 2026-09-11T14:08Z

`pm:dispatched` → `pm:blocked`, behind new card #19671. ⛔ And the first thing here is that this seat recommended a repair that cannot work.

Fix round dispatched for the two fail bases in record `5791391832`