[finding] `DeleteMetaItemResponseSchema` declares neither `seq` nor `projectionApplied`, which its branch sends — the two sibling verbs on the SAME door declare both, and the save side got that by ruling (#5745)

[os-litant]

Measured by the #13023 dev (PR #13148) while binding meta.deleteItem's return to the exported spec type. Filed by the domain:cli execution seat (#6024, session session_01UjujZN219uFzBhSYfMykCd) on its behalf — that seat's filing channel returns HTTP 403 and its permitted dedupe fallback was rate-limited, so it handed the measurement over rather than mint a possible duplicate. I ran the dedupe search and re-verified every reading below against origin/main before filing.

The asymmetry

The metadata door has three write verbs. All three run the same ADR-0094 mutation projector and the same history append, and all three put seq (and projectionApplied, when a projector is registered) on the wire. Two declare them. The third does not.

verb	response schema	declares seq / projectionApplied
save	SaveMetaItemResponseSchema — packages/spec/src/api/protocol.zod.ts:611	✅ both
publish	PublishMetaItemResponseSchema — :774	✅ both
delete / reset	DeleteMetaItemResponseSchema — :1197	⛔ neither

The third schema, in full:

export const DeleteMetaItemResponseSchema = lazySchema(() => z.object({
  success: z.boolean(),
  reset: z.boolean().optional(),
  message: z.string().optional(),
}));

And what its own branch really returns — packages/metadata-protocol/src/protocol.ts, the repository-delete branch:

return {
    success: true,
    reset: true,
    seq: result.seq,
    ...(deleteProjection ? { projectionApplied: deleteProjection } : {}),

deleteProjection is the awaited runMutationProjector({ …, state: 'deleted' }) immediately above it — the same projector call the other two verbs make.

⭐ Why this is more than "one more undeclared key"

The save side was filled in by a ruling, and this one was not. #5745 — "[#5563 附带裁决] SaveMetaItemResponseSchema 补齐实现实际返回的字段(version / seq / state / projectionApplied)" — closed exactly this gap for saveMetaItem, naming seq and projectionApplied among the fields to add.

⇒ So this is not an oversight nobody has looked at. It is a decision that was made for one verb and never carried to its sibling, which is why the three verbs now answer the same question three different ways while doing the same work.

What it costs

projectionApplied is the channel a caller uses to learn whether the ADR-0094 projector actually succeeded — the neighbouring schemas' own prose says a caller should read projectionApplied.success rather than rely on the 200. On the reset path that channel is undeclared, so a first-party caller cannot reach it without an as any, and a generated client will not carry it at all.

seq is the ordering token the history/audit trail is read by. Undeclared here, declared on both siblings.

⚠️ And it now has a second-order effect, freshly created: PR #13148 binds meta.deleteItem's return to this schema (per the #13023 ruling — the spec is the contract). That is correct and should land, but it means the SDK's declaration is now a genuine narrowing relative to the wire on exactly these two keys. Closing this card widens the schema and the SDK inherits the fix for free, because it binds the type rather than transcribing a member list. ⛔ Nothing in packages/client needs to change for that; do not re-open #13023.

Scope note, not decided here

Both keys would be .optional() — seq is absent on the branches that do not touch a repository row, and projectionApplied only when a projector is registered (the same conditional the siblings already declare). Whether to mirror the siblings' shapes exactly, or to declare the narrower thing this branch actually sends, is the decision; this card does not take it.

⛔ This card claims nothing about the wire being wrong. The wire is right and the two siblings agree with it. It is the third declaration that is short.

Related: #5745 (the sibling's own fix, by ruling), #11679 (the same declared-vs-sent gap on DeleteMetaItemRequest), #13023 / PR #13148 (the SDK-side binding that surfaced this).

[os-trump]

Claim: PM loop round R7 (domain:spec seat)
Session: session_01LpRNHxWZgSUgVnFT9mQQo4
Branch: claude/issue-13155-delete-response-parity
Worktree: objectstack-issue-13155
Domain: domain:spec
File surface: packages/spec/src/api/protocol.zod.ts (DeleteMetaItemResponseSchema at ~:1197), its pin/test coverage in the same package, and the regenerated api-surface / export-origins / reference pages the gates prove stale. Stop on breach; explain in the report.
Container & model: M, mode:subagent, model: claude-opus-5 — Clause-② both legs (PATH: packages/spec/src/**; CONTENT: a published response accept set widens). Fable is mandatory for this shape and is measured unavailable (account-level 429 twice today), so this runs at opus under the quota-exhaustion exemption (maintainer 2026-08-13, verbatim 「fable 如果用完了,可以用 opus」), floor opus, ⛔ no lower. Compensating control unchanged: needs:contract-review attaches to the PR and STAYS; review is parked for the chain.
PM ruling of the open decision (the card explicitly leaves it — recorded here for the veto window): mirror the siblings exactly, ⛔ not a bespoke narrower shape. Grounds: the defect's own diagnosis is asymmetry — three verbs on one door, doing the same work, answering the same question three ways — and a fourth distinct shape does not cure that; #5745 already ruled the shape for saveMetaItem, so carrying it to its sibling applies an existing ruling rather than minting a new one; and both keys are .optional() on the siblings for exactly the conditions that hold here (seq absent on branches that touch no repository row, projectionApplied only when a projector is registered), so the two candidate shapes most likely coincide anyway. Premise: the delete branch's sent set is expressible in the siblings' shape. ⛔ Premise fails — the branch sends something that shape cannot express, or mirroring would declare a key this branch provably never sends ⇒ report the fork on this card, never force the mirror.
Fences:

• ⛔ Do not edit packages/spec/src/type-alias-convention.pin.test.ts — the parked PR Retire the paper metadata-customization protocol with its full coupling set #13186 owns that file and a collision reds the single-writer-path guard. If the change genuinely requires it, STOP and report; the card then waits for Retire the paper metadata-customization protocol with its full coupling set #13186 to land rather than racing it.
• ⛔ Nothing in packages/client changes and meta.deleteItem declares a return type — { type, name, deleted } — that the reset door has never answered #13023 is not re-opened — PR fix(client): declare the response the meta reset door actually sends on both deleteItem twins #13148 binds the type, so it inherits this fix for free (the card says so and it is correct).
• ⛔ No claim that the wire is wrong: the wire is right, the two siblings agree with it, the third declaration is short.
  Serial constraints cleared, checked rather than assumed: api/protocol.zod.ts is free — the parked fix(spec): widen AnalyticsResultResponseSchema and TriggerFlowResponseSchema data to producer-contract parity #13184 (Two packages/spec route response schemas declare a data NARROWER than the contract their route relays — AnalyticsResultResponseSchema and TriggerFlowResponseSchema #13078) considered it and deliberately left its module-local alias alone, so it is not in that PR's file list (verified against the branch diff). In-flight [finding] FormFieldPublicPickerSchema tells authors the picker target resolves from referenceTo — a key FieldSchema rejects — and that sentence ships into generated JSON Schema and docs #13138 (ui/view.zod.ts) and [finding] 1316 duplicate heading anchors remain on 162 of 214 reference pages — the section grammar's own headings (### Properties, ### Allowed Values, #### Option N) carry no qualifier at all #12862 (scripts/lib/schema-section.ts, lib/format-type.ts) are disjoint. Shared regenerated artifacts (api-surface/api.json, export-origins/api.json) are also touched by the parked fix(spec): widen AnalyticsResultResponseSchema and TriggerFlowResponseSchema data to producer-contract parity #13184 — they serialize at the landing window, and the dev merges current main before its final push.

---

Generated by Claude Code

[os-trump]

Dev seat claiming this card.

• Session: session_01LpRNHxWZgSUgVnFT9mQQo4
• Branch: claude/issue-13155-delete-response-parity
• Dispatched by the domain:spec PM seat, loop R7.

Starting the premise check on origin/main before any edit.

---

Generated by Claude Code

[os-trump]

os-dev-report

{
  "issue": 13155,
  "status": "done",
  "branch": "claude/issue-13155-delete-response-parity",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/13208",
  "premise_still_valid": true,
  "summary": "DeleteMetaItemResponseSchema now declares seq and projectionApplied, mirroring the two sibling verbs per the PM ruling; the widening carries #5745's decision to the third verb rather than minting a new shape. Premise VERIFIED on origin/main before editing: the sole producer is deleteMetaItem, the REST door does res.json(result) verbatim so the protocol return IS the wire body, MutationProjectionOutcome is byte-identical to the inline object both siblings declare, and DeleteResult.seq is the same integer sequence. One measured deviation from a literal member-copy, and it is what the premise fence exists for: deleteMetaItem has FOUR success returns and only one appends a history event, so both keys land .optional() here while seq is REQUIRED on both siblings — the two no-op returns and the control-plane legacy raw-engine delete (#5264) all answer without it, and declaring it required would make three of the producer's own returns fail their own contract. version and advisories are deliberately NOT declared and that is pinned, because this branch provably never sends them. FENCES HELD: type-alias-convention.pin.test.ts was NOT edited — the widening mints no new pin (.optional() widens input and output identically, the precedent that file already records for advisories), confirmed mechanically by check:spec-parsed-alias reporting 837 pinned isomorphic, exactly the count that file asserts. Nothing in packages/client changed and #13023 was not re-opened. No claim is made anywhere that the wire is wrong. PR is DRAFT with needs:contract-review attached (label write confirmed by read-back as size/l + needs:contract-review, size/l preserved); never marked ready, never enqueued, auto-merge never armed.",
  "tests": "Union run AFTER the final commit, on c2b36e4a3 (current origin/main merged first; closure reinstalled and rebuilt since packages/spec moved on the incoming side). Exit codes captured before any pipe; verdict lines quoted from each gate. check:generated: 'All 14 generated artifacts are up to date' (check:api-surface and check:export-origins were already green — the change adds no export; gen:docs and gen:strictness-ledger were regenerated by --fix on exactly the two proved stale, never by hand). spec typecheck: 'check:test-typecheck: OK — @objectstack/spec's test layer compiles under packages/spec/tsconfig.test.json'. spec suite: 443 files / 11775 tests passed. Three door conformance suites in objectql: 3 files / 28 tests passed. Client suite (corroborating, nothing should change): 30 files / 402 tests passed — NOTE its first run showed 7 files failing to LOAD with 'Failed to resolve entry for package @objectstack/runtime', an unbuilt-closure resolution error and NOT a red gate; rebuilt and re-run as a real measurement. Whole-repo pnpm lint (eslint . --no-inline-config): exit 0, no output — so no narrowing is claimed. check:objectql-double-limit, check:engine-double-contract, check:where-matcher, check:cross-package-test-inputs, check:test-source-alias, check:type-check-coverage, check:query-options-erasure, check:doc-authoring, check:merge-driver, check:published-files and the six changeset-family gates all exit 0. Gate family derived with 'node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack', no hand-fed paths. ABLATION, rebuild-backed both legs because objectql→spec is registered UNALIASED in KNOWN_UNALIASED_TEST_IMPORTS (resolves through dist/): attempt 1 was a FALSE GREEN and is reported as such — narrowing the schema alone is a bare deletion against the regenerated authorable-surface/api.json, so gen:schema's deletion guard failed the build, the marker never reached dist/, and the suite ran the PRE-MUTATION build and passed 5/5; ablation-dist-preflight caught it ('marker ABSENT from dist/'). Attempt 2 reverted the baseline lines in the same mutation: mutation confirmed on disk by anchored counts (injected marker = 1, removed describe anchor = 0, baseline seq lines = 0, baseline proj lines = 0), build exit 0, preflight 'marker present in 2 built files', then ABLATED_TEST_EXIT=1 with 4 failed / 1 passed and the expected diagnostics 'expected [ seq ] to deeply equal []' and 'expected [ seq, projectionApplied ] to deeply equal []'. The single case green under ablation is the no-overlay-row no-op, which carries neither key, so nothing is stripped there either way. RESTORE leg symmetric and verified, not assumed: sources restored with 'git checkout HEAD --', git diff HEAD empty, both blob hashes equal their HEAD blobs, spec rebuilt, preflight --absent reported 'marker absent from all 215 built files', suite green again 5/5. NOT MEASURED, stated rather than passed off as a pass: packages/objectql's typecheck is tsc --noEmit against a tsconfig whose exclude carries the test glob and the package has no test-layer program or debt ledger, so tsc --listFiles returns 0 hits for the new conformance file — its types are exercised by the vitest run and by CI, not by a static program. A pre-existing package property, not something this PR changes. By contrast the spec-side type pins WERE verified real rather than phantom: tsc -p tsconfig.test.json --listFiles lists protocol.test.ts.",
  "mcp_calls": "10 — for the whole run (claim comment, issue body, issue comments, create PR, one failed get_labels on a PR number, two pull_request_read, one label write, this report comment, and its read-back). Repo-scoped REST is 403 for this seat (measured: gh absent, api.github.com/repos/... returns 403 'GitHub access is not enabled for this session'), so the label write took the declared MCP fallback — read current set, union, whole-set write, compare read-back — rather than the additive REST endpoint. PR 13184's changeset was read via git fetch of refs/pull/13184/head, at zero MCP cost.",
  "open_questions": [],
  "out_of_scope_findings": []
}

No out-of-scope issues were filed, and that is a finding rather than an omission: the two candidates both turned out to be already-tracked or not defects. (1) packages/objectql hiding its test layer from every tsc program is real, but check:type-check-coverage already accounts for it — 18 package(s) still hide their own tests from tsc ... 1188 frozen raw errors in TEST_DEBT — so it is a ledgered, shrink-only condition with an owner, not an unrecorded gap. (2) The ablation false-green is a methodology result, not a repo defect; it is written up in the PR body because it is genuinely useful — the two declared keys are now guarded by two independent mechanisms, so a future bare deletion of either cannot land quietly.

One in-scope repair worth flagging for review: check:objectql-double-limit failed my new conformance file as a NEW limit-blind find double (the two sibling conformance files predate that gate and sit in its shrink-only baseline). Fixed inside my own new file per the gate's own prescription — the bound is applied after the filter, by presence — not by touching the baseline.

---

Generated by Claude Code