spec(ai): declare the build-progress PHASE vocabulary the cloud agent loop emits during post-apply verification (cloud#2172 ruling A)

[hotlong]

Filed by the director seat (summon #24, session_01Wj1HUjzyeiBQ8atRf1ZhaL) executing the maintainer's ruling on objectstack-ai/cloud#2172 — batch #141 item 5, letter A, reply verbatim 「同意」. ⛔ No domain:* set by the filer; triage assigns the lane (the change lands in packages/spec, so the anchoring rule points at domain:spec).

Dedup terms: build progress phase vocabulary · data-build-progress · post-apply verification · verify phase · progress frame

What is being declared

A closed enum of build-progress phases that the cloud AI-studio agent loop emits as data- SSE frames during the post-apply verification window (the ~111 s / 9-tool-call stretch after apply_blueprint / apply_edit returns, measured on cloud#1838), and that the objectui chat panel renders (objectui#7388 block 2). Today the encoder passes data- parts generically and objectui has a generic data-* receiver; what is missing is the vocabulary, so an unknown value can be warned about instead of silently rendering the previous label.

The ruling fixed three things (cloud#2172, ruling comment): sender = the agent loop; channel = the existing data- frames; vocabulary = @objectstack/spec. This card is the third.

What this card decides and what it does not

• Decides: the phase enum's members, its zod schema and export, the frame's minimal shape (phase + optional hop counter / tool name), and the docblock stating who emits it and who consumes it.
• ⛔ Does not decide the UI copy for each phase (that is objectui#7388's), nor the emitter's placement in the cloud agent loop (that is cloud#2172's).
• This is a new published surface on packages/spec ⇒ the claim declares Clause-②: yes and the change goes through the seat's in-seat contract review.

Acceptance

1. The enum is exported from packages/spec with a docblock naming producer (cloud agent loop, post-apply verification) and consumer (objectui chat panel).
2. A pin: an unknown phase value is refused by the schema (loud), a known one parses; the data- frame shape parses with and without the optional fields.
3. Changeset: minor on @objectstack/spec (additive public surface).

Links

• Ruling: objectstack-ai/cloud#2172 (batch Fix query tests after schema migration to object-based filter syntax #141 item 5)
• Emitter (cloud): objectstack-ai/cloud#2172 — its emitter may land behind this card's Blocked-by:
• Receiver (objectui): plugin-chatbot: the build-progress panel cannot express a verification phase — an unknown phase silently reads as 「Building…」, and its copy is hardcoded English objectui#7388 block 2 — takes Blocked-by: this card
• Origin of the measurement: objectstack-ai/cloud#1838

Refs objectstack-ai/cloud#2172 · Refs objectstack-ai/objectui#7388

---

Generated by Claude Code

[os-sam]

Triage: routed and graded — domain:spec · p2 · Feature

Triage seat (objectstack#6015), session_01U6nJJSVZb476ckbRxmaMCA, R+251, 2026-09-16T16:06Z. ⛔ This seat does not claim, dispatch or write code.

Filed by the director seat executing the maintainer's ruling on objectstack-ai/cloud#2172 (batch #141 item 5, letter A, reply verbatim 「同意」), with the lane left to triage.

• domain:spec — the change lands in packages/spec; the anchoring rule points here and the filer said so. ⭐ Confirmed on merits rather than inherited: a closed enum declared in the spec package is the spec lane's whoever consumes it, per 「新 packages/spec 工作恒由 domain:spec 席收口,不论谁需要它」 — cloud emits it and objectui renders it, and neither fact moves the lane.
• type: Feature — 「扩大接受集/公开面 ⇒ Feature」. This declares a new vocabulary that does not exist today. ⛔ Not a Bug: nothing is currently violating a contract; there is no contract yet, which is the point.
• priority:p2 — it is a ruled deliverable with a named consumer already emitting the frames, so the vocabulary is the missing half of shipped behaviour rather than speculative surface. ⛔ Not p1: the cloud loop works today and nothing is broken without it.

⚠️ For the claiming seat: declaring a new published enum is unambiguously accept-set growth ⇒ Clause-②: yes, and the 入队闸门 will read the claim comment's row. ⭐ Note also the ruling names a closed enum — 「先量再写」 applies to its members: they should come from the frames the loop actually emits (the ~111 s / 9-tool-call post-apply window measured on cloud#1838), ⛔ not from what a phase vocabulary could contain.

_{本评论来自分诊座位}

---

Generated by Claude Code

[os-warren]

Claim: PM loop, domain:spec execution seat
Session: session_01KB5PFtxuy1x3dcR5gxudx6
Branch: claude/issue-18451-build-progress-phase-vocabulary
Worktree: objectstack-issue-18451
Domain: domain:spec
File surface: packages/spec/src/ai/** (the new module + its index.ts export), packages/spec/src/contracts/ai-service.ts docblock if the frame example needs to name the enum, the generated api-surface / export-origins / reference-page artifacts that follow from a new export, and .changeset/. Stop on breach; explain in the report.
Container & model: M, mode:subagent, model: claude-opus-5 — the dispatch tier, measured this shift by transcript census on this seat's os-dev runs (138 of 138 "model":"claude-opus-5" on the most recent), ⛔ not from a self-report.
Clause-②: yes
Thread-read: 5700618838
Serial constraints cleared: none. All 17 open PRs' /files were enumerated at 2026-09-16T18:48:40Z and zero touch packages/spec/src/ai/** or packages/spec/src/contracts/ai-service.ts; the only packages/spec/src/contracts/ file in flight is email-service.ts (PR #18482).

Premises measured on origin/main fb6b2c369e, 2026-09-16T18:51:25Z — ⛔ not restated from the card

reading	value
the data- frame channel the vocabulary attaches to	exists in prose already — packages/spec/src/contracts/ai-service.ts:639 and :644 document data-build-progress and its example carries a phase
a declared phase vocabulary	absent — the two hits above are the only build-progress/buildProgress/BuildProgressPhase occurrences under packages/spec/src
lit control, same corpus	z.enum( in packages/spec/src/ai/ → agent.zod.ts 5, conversation.zod.ts 6, embedding.zod.ts 2 ⇒ the probe can see enums where they exist
action surface	git log -8 origin/main -- packages/spec/src/ai/ — last touch d34f9b6cc2 (#17699), nothing pending on this module

⇒ the card's premise holds: the channel ships, the vocabulary does not.

⚠️ Scope reality the dev inherits: the ruling this card executes lives on objectstack-ai/cloud#2172, and the cloud repository is not reachable from this seat (session repo scope is objectstack + objectui). The ruling text is therefore taken from the director seat's filing on this card, verbatim, and ⛔ the dev must not report the cloud issue as "checked" — an unreachable repo is a reading gap, not a clean read. objectui#7388 is reachable and may be read.

The enum is closed by the ruling, so its membership is measured before it is written, not designed here. Nothing in this claim decides the members.

---

Generated by Claude Code

[os-warren]

os-dev-report

{
  "issue": 18451,
  "status": "done",
  "session": "session_01KB5PFtxuy1x3dcR5gxudx6",
  "branch": "claude/issue-18451-build-progress-phase-vocabulary",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18528",
  "head_sha": "3233eea6733eb621d76ed7386c558e290c987497",
  "premise_still_valid": true,
  "summary": "Declared the closed build-progress PHASE vocabulary on `@objectstack/spec/ai` (new module packages/spec/src/ai/build-progress.zod.ts, exported from the ai barrel): BUILD_PROGRESS_PHASES / BuildProgressPhaseSchema / BuildProgressPhase, BuildProgressFrameSchema / BuildProgressFrame, and BUILD_PROGRESS_FRAME_TYPE. api-surface reads 6 added / 0 removed on ./ai - purely additive, matching the `minor` changeset and `Clause-②: yes (widening)` (the arm is the spelling check-adr-0087-registration's own self-test blesses for a widening minor; origin/main AGENTS.md L1067 is the governing sentence, and I read it from the `git show origin/main:AGENTS.md` blob, NOT the shared checkout, whose L1043 carries the older phrasing). The frame is deliberately z.looseObject: the panel fields objectui already reads ride the same frame, so a strict schema would refuse every frame shipping today. ALL THREE PREMISES RE-MEASURED MYSELF AND HELD. READING GAP, stated plainly and NOT papered over: the ruling lives on objectstack-ai/cloud#2172 and the `cloud` repo is NOT reachable from this session. The ruling text was taken from the director seat's filing on the card, verbatim; I do NOT report the cloud issue as checked. objectui#7388 IS reachable and I read it in full plus all 10 of its comments - it supplied the consumer half, and its history independently confirms the emitter fact this card's ruling settles (a returned tool cannot emit on its own ctx.onProgress, which is exactly why the producer is the agent LOOP and why objectui#7388's own piece 2 stalled). FILE SURFACE: see `file_surface_accounting` - 17 files in three buckets, bucket 3 (my own judgement calls) is EMPTY. One earlier discretionary edit (a table row in quick-reference.mdx) was dropped on the seat's instruction, reducing that file to a one-line count correction. contracts/ai-service.ts was deliberately NOT edited. I attached NO labels: needs:contract-review is the seat's on both carriers.",
  "enum_membership_sourcing": {
    "note": "The ruling names a CLOSED enum, so membership was measured before it was written. Per-member provenance is also recorded in the module source and in the PR body.",
    "structure": "objectui `packages/plugin-chatbot/src/ChatbotEnhanced.tsx:163` - the consumer's own declared union `phase is structure | data | done`; and `packages/plugin-chatbot/src/mapMessages.ts:744` - `const phase = d.phase === 'data' || d.phase === 'done' ? d.phase : 'structure'`, where it is also the coercion default. Read in the objectui checkout at ff1d5ea8d171b65ed5576199807c382a3ccc5b49.",
    "data": "same two sites: ChatbotEnhanced.tsx:163 and mapMessages.ts:744 (objectui @ ff1d5ea8d1).",
    "done": "same two sites: ChatbotEnhanced.tsx:163 and mapMessages.ts:744 (objectui @ ff1d5ea8d1).",
    "verify": "objectui#7388 body, verbatim: 'Admit a `verify` phase in `extractBuildProgress` and give it its own line in the progress panel' - the post-apply verification window the card exists to name. Corroborated IN THIS REPO by packages/spec/src/system/constants/platform-tool-names.ts, whose `service-ai-studio` group registers the `verify_build` tool and whose docblock calls that group the build product's 'metadata/blueprint/package/verify tools'.",
    "could_not_source": "The PRODUCER's own emission list. cloud#2172 and cloud#1838 are unreachable from this session, so no member was taken from what the loop is observed to emit. Nothing was rounded out to cover that gap: the enum is exactly the four members above. If the cloud loop turns out to emit a fifth phase, that is an additive follow-up with a real source, not something I should have guessed here.",
    "explicitly_rejected": "`designing` - it belongs to the SIBLING `data-blueprint-progress` frame (ChatBlueprintProgress, `d.phase === 'done' ? 'done' : 'designing'`), a different channel with its own reader. Folding it in would be the 'one enum for two frames' mistake, and a test asserts it does NOT parse here."
  },
  "file_surface_accounting": {
    "total_files": 17,
    "declared_in_claim_4": [
      ".changeset/build-progress-phase-vocabulary.md",
      "packages/spec/src/ai/build-progress.zod.ts",
      "packages/spec/src/ai/build-progress.test.ts",
      "packages/spec/src/ai/index.ts"
    ],
    "bucket_1_generated_10": {
      "proof": "NOT asserted - demonstrated. All ten were reverted to the merge base fb6b2c369e (the new page deleted) and re-produced by re-running the generators on that tree: `check:generated --fix` wrote gen:schema, gen:api-surface, gen:export-origins, gen:docs, gen:strictness-ledger; gen:declaration-map was then run explicitly, because the aggregate does not flag declaration-map stale when its companion export-origins is reverted in the same stroke (worth knowing on its own). `git diff HEAD --stat` over all ten came back EMPTY - byte-identical reproduction. The tree was then restored with `git restore --source=HEAD --staged --worktree .` (index AND tree, because `git checkout BASE -- path` stages what it retrieves) and `git status --porcelain` verified empty.",
      "files": {
        "packages/spec/api-surface/ai.json": "gen:api-surface",
        "packages/spec/export-origins/ai.json": "gen:export-origins",
        "packages/spec/declaration-map/ai.json": "gen:declaration-map",
        "packages/spec/authorable-surface/ai.json": "gen:schema (via check:authorable-surface)",
        "packages/spec/json-schema.manifest/ai.json": "gen:schema",
        "docs/audits/2026-07-unknown-key-strictness-ledger.counts.md": "gen:strictness-ledger",
        "content/docs/references/ai/build-progress.mdx": "gen:docs (new page)",
        "content/docs/references/ai/index.mdx": "gen:docs",
        "content/docs/references/ai/meta.json": "gen:docs",
        "content/docs/references/index.mdx": "gen:docs"
      }
    },
    "bucket_2_required_by_a_gate_3": {
      "packages/spec/src/type-alias-convention.pin.test.ts": "GATE `pnpm check:spec-parsed-alias`, RED before the edit, quoted: '`BuildProgressFrame` is the AUTHOR state of `BuildProgressFrameSchema` and nothing names its PARSED state. Declare `export type BuildProgressFrameParsed = z.infer of BuildProgressFrameSchema` next to it so a consumer holding a parse result has a name - or, if `z.input` and `z.infer` of `BuildProgressFrameSchema` are the same type, pin it in packages/spec/src/type-alias-convention.pin.test.ts instead.' It also red on the sibling: '`BuildProgressPhase` names the PARSED state ... ADR-0122 reserves the bare name for the AUTHOR state.' Both schemas ARE isomorphic, so the pin route is the one ADR-0122 prescribes, and that file's own header says why it beats the alternative: 'a permanent synonym is a name an author can only pick wrongly.' The file is ALSO the gate's machine-readable exemption registry ('scripts/check-spec-parsed-alias.mjs reads THIS file to decide which bare aliases are allowed to have no XParsed'), so the registration cannot live anywhere else. Edit: +1 module import slot M188, 2 Iso pins (Iso875/Iso876), and the pin count 783 -> 785 in all three places that state it (header prose, case title, runtime assertion - a companion test asserts all three agree). tsc is what PROVES the two new pins.",
      "packages/spec/llms.txt": "GATE `pnpm check:llms-txt`, RED before the edit, quoted: 'packages/spec/llms.txt:95 [count] domain `ai` declares 11 schemas; src/ai/ holds 12' and 'packages/spec/llms.txt:80 [count] heading declares 200 schemas; packages/spec/src/ holds 201'. Hand-kept BY DESIGN - the gate says 'there is deliberately no gen:llms-txt' - and it ships to AI consumers inside the npm tarball, so a stale claim there becomes generated code that does not compile. Edit: 11 -> 12, 200 -> 201, and the domain's representative subject list gained the new module per the gate's own instruction to 'correct the number AND the prose'. Green after: '97 claim(s) re-derived ... 14 domains, 201 schemas'.",
      "content/docs/getting-started/quick-reference.mdx": "GATE `pnpm check:quick-reference-counts`, RED before the edit, quoted: 'quick-reference.mdx:111 [total] section \"AI Protocol\" declares \"of 11 schemas\" but content/docs/references/ai/ publishes 12 page(s)'. REDUCED TO THE GATE-REQUIRED MINIMUM on the seat's instruction: now a single line, '11 of 11' -> '11 of 12' (+1/-1). An earlier revision of this branch also added a table row for the new reference page; that was bucket-3 polish, not needed for correctness, and it was dropped in commit 3233eea673. The section now carries the same partial shape the API section already has at '17 of 31'."
    },
    "bucket_3_my_own_judgement_calls": "NONE. Every non-declared path is either generator output reproducible on a clean tree (bucket 1) or named by a gate together with its remedy (bucket 2). If that reads as the claim's surface wording having been narrower than what one new exported schema mechanically forces in this tree, that is the seat's to record, per the seat's own note."
  },
  "tests": "ALL on final head 3233eea673 with a clean tree, after `pnpm --filter @objectstack/spec build`. PREMISES RE-MEASURED (not restated): (1) contracts/ai-service.ts:639 and :644 carry `data-build-progress` and the phase example - confirmed, exactly those two lines. (2) dark probe `git grep -nE 'build-progress|buildProgress|BuildProgressPhase' -- packages/spec/src` returned exactly 2 lines, both that prose, so no vocabulary existed; LIT CONTROL same corpus, `z.enum(` occurrences via `grep -o | wc -l` = agent 5, conversation 6, embedding 2 (the seat's three) and `git grep -c` shows 7 files under src/ai carry it (also mcp 2, model-registry 4, skill 3, solution-blueprint 13) - the probe sees enums where they exist, so the zero is real. (3) `git log -8 origin/main -- packages/spec/src/ai/` ends at d34f9b6cc2 (#17699) - confirmed. NEW PIN packages/spec/src/ai/build-progress.test.ts: 16 tests, exit 0. AFFECTED PACKAGE `pnpm --filter @objectstack/spec test` under the shared lock: 'Test Files 484 passed (484)', 'Tests 13789 passed (13789)', VERDICT command-exit 0. TYPECHECK `pnpm --filter @objectstack/spec typecheck` VERDICT command-exit 0. COVERAGE CAVEAT measured with --listFiles: the main `tsc --noEmit` program does NOT include **/*.test.ts (0 hits for build-progress.test.ts and for type-alias-convention.pin.test.ts, 1 hit for the source); the test layer runs under tsconfig.test.json, whose raw tsc exits 2 with 259 LEDGERED pre-existing errors - a redirect-then-$? grep for my files over that output returns ZERO (grep exit 1) with a lit control finding 5 in conversation.test.ts. That zero is what proves the two new ADR-0122 Iso assertions. DERIVED GATES `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` = 108 families. The list was derived four times - before and after two origin/main advances, and again after my later added paths - and was IDENTICAL every time. All 108 RUN on this final head, each exit code captured by redirect-then-$?, never through a pipe, then reconciled with `--ran`: '108 derived, 108 run, 0 NOT-MEASURED, 0 UNRUN' (a DERIVED zero - every family carries a code and none is 3). 107 exit 0. THE ONE NON-ZERO IS NOT THIS DIFF: `pnpm check:cross-package-test-inputs` exits 1 naming packages/cli/test/init-created-files-summary.e2e.test.ts descending packages/spec/dist/ - paths my diff does not touch. Three-leg control on a throwaway worktree at pristine origin/main 97233b90ca with ZERO of my changes: no dist -> exit 0; same tree with only my built packages/spec/dist copied in -> exit 1 with the byte-identical finding; dist removed -> exit 0. Filed below. FIVE GATES were initially PREREQUISITE-NOT-MET (two exit 3, plus exit-1s whose own text said nothing was measured). Rather than report them NOT MEASURED I built @objectstack/lint, @objectstack/formula, @objectstack/client-react and then the whole workspace (72/72 tasks); all five then exit 0 - check:doc-formula-expressions, check:doc-security-posture, check:skill-examples, check:dual-build-cjs-loads, and check:type-check-debt (re-measured 5 ledger entries, 55 raw errors, none above its recorded number). REPO-WIDE `pnpm lint` (eslint . --no-inline-config) exit 0 - run IN FULL, not narrowed, so none of the three narrowing-evidence pieces is owed. `check:generated` exit 0: all 15 artifacts up to date; authorable-surface.base.json was NOT moved. `check:nul-bytes` exit 0, plus a manual control-byte scan of every file I wrote (grep -naP over the C0 range, exit 1 = clean). ABLATION (from the COMMITTED state; each leg proves the mutation reached disk BEFORE any verdict is read - occurrence counts of BOTH the removed and the injected text, plus a blob hash differing from the HEAD blob 53034bc672c8de6dbe800cd923bba8b7f237c66d - and restores by OBSERVED STATE via `git checkout HEAD -- path`, hash back to the HEAD blob and `git diff HEAD` empty, never by exit code; trap on EXIT/INT/TERM with absolute paths, empty hash treated as FAILURE). No dist preflight was owed: the test imports the module by relative source specifier, not through the package exports. LEG 1 z.enum(BUILD_PROGRESS_PHASES) -> z.string(): removed-text 0 / injected 1, blob 69eb192cc1, run exit 1, 3 failed | 13 passed - all three refusal cases red BY NAME, every known-phase control still green. LEG 2 z.looseObject -> z.strictObject: removed-text 0 / injected 1, blob 54521efa6f, run exit 1, 1 failed | 15 passed - exactly the FLOOR case. Both legs restored and verified; final `git status --porcelain` empty. Direction observed = turn-red, as predicted. CI on 3233eea673 after the last push, latest run per check NAME: 6 success (Check Documentation Links; Flag docs affected by code changes; Governed Surface Queue Guard; No other open PR may claim the same issue; Spec property liveness; filter), 22 in_progress, 4 skipped (NOT verdicts: Auto Label, Check PR Size, Console Pin Gate, Packed-tarball smoke), ZERO failures so far. CI convergence is the PM's, not mine - I did not wait. READING `node scripts/pm/check-clause2-carriers.mjs --pair 18528` :: exit 0 - both carriers agree (reported as a reading, not a blocker, and I attached no label myself). PR body verified by read-back after every write; final state is one footer, `Fixes #18451` first line, `Clause-②: yes (widening)` line-initial, still DRAFT.",
  "mcp_calls": "0 - no MCP tool of any kind was called this run. Every GitHub read and write went through the REST proxy with curl and $GITHUB_TOKEN. Tool names called: (none).",
  "api_writes": "4 REST proxy writes. (1) POST /repos/objectstack-ai/objectstack/pulls - draft PR #18528. (2) PATCH /repos/objectstack-ai/objectstack/pulls/18528 - body rewritten with the three-bucket file accounting and the new head sha, after the seat's mid-task note and after dropping the discretionary table row, so the body would not describe a change the diff no longer contains. (3) PATCH /repos/.../pulls/18528 again - write 2 came back with TWO footers: the platform APPENDED its own bare block (blank line, rule, bare footer) on top of the session-URL footer I sent, which is the measured edit-channel behaviour AGENTS.md warns about. Write 3 re-sent the body with NO footer and the session named in body PROSE instead; read-back confirms exactly one footer. (4) POST /repos/objectstack-ai/objectstack/issues/18451/comments - this report. ZERO label writes - needs:contract-review belongs to the seat on both carriers, per the dispatch. Separately, 5 `git push` calls (the empty-branch routing probe plus four commits), which are not REST writes. One earlier POST /pulls attempt returned HTTP 415 for a missing Content-Type header and created nothing.",
  "open_questions": [],
  "out_of_scope_findings": [
    "to file (3 classes, dedupe words: `cross-package-test-inputs dist presence` · `gate verdict depends on gitignored build output` · `check:cross-package-test-inputs conditional green` · `spec/dist walk undeclared glob` · `init-created-files-summary e2e walk`): CLASS (a) reproducible defect - `pnpm check:cross-package-test-inputs` reaches OPPOSITE verdicts on the same commit depending on whether the gitignored packages/spec/dist/ happens to exist. Repro, all on pristine origin/main 97233b90ca with no source change: without dist exit 0; with a built dist copied in exit 1, naming packages/cli/test/init-created-files-summary.e2e.test.ts and packages/spec/dist/; with the dist removed again exit 0. This is the 'verifier that silently degrades' shape Route and surface ownership rule 3 names: green in the state CI may run it in, red for anyone who followed AGENTS.md's instruction to build before trusting a dist-reading gate - so the blind spot it exists to close (#7802) is itself conditional on local build state. Remedy is the gate's own printed option 1 or 2 (narrow the walk, or declare the root and name the test in heldBy). Not filed by me: dev does not POST /issues.",
    "noted, not filed: objectui's sibling reader `extractBlueprintProgress` collapses an unknown `data-blueprint-progress` phase to 'designing' by exactly the pattern this card fixes on the build channel. Deliberately not touched and deliberately not filed - objectui#7388's PM ruling already examined it and found the risk asymmetric (its docblock makes 'only done is authoritative' a contract, and its panel is superseded by the authoritative card the moment propose_blueprint lands), and its vocabulary has no producer ruling. Carrier: whoever picks up a data-blueprint-progress vocabulary card, if one is ever ruled; absent that, carrier: none.",
    "noted, not filed: `check:generated --fix` does NOT flag packages/spec/declaration-map/ai.json stale when its companion export-origins/ai.json is reverted in the same stroke - the two are compared against each other, so a consistently-stale PAIR passes the aggregate. I only saw it because the bucket-1 reproduction reverted both at once, and `gen:declaration-map` then reproduced my bytes exactly. Not filed: I could not show it produces a wrong verdict on any real workflow (a normal edit moves one side first), so it is an observation about a detection order, not a demonstrated defect. Carrier: whoever next changes how declaration-map staleness is computed.",
    "noted, not filed: the regenerated packages/spec/authorable-surface/ai.json now lists `ai/BuildProgressFrame:phase|hop|tool` and json-schema.manifest/ai.json lists the two new schemas. Reviewers should not read that as a claim the frame is author-writable metadata - the generator enumerates every schema in the domain (`ai/CodeContent:language` sits there on the same basis). Pre-existing generator behaviour, not a change this PR makes. Carrier: none.",
    "noted, not filed: `contracts/ai-service.ts`'s `onProgress` docblock still describes the `data-build-progress` frame in prose without naming the new enum. The claim permitted editing it; I left it untouched to keep this diff off a contracts/ path while PR #18482 is in flight there, and because the new module's own docblock points back at it. Carrier: a follow-up that cross-links the two, or the cloud#2172 emitter card when it lands."
  ]
}

---

Generated by Claude Code

[os-warren]

Round returned done — collected. Bucket 3 is empty, and the narrow surface was the seat's error

domain:spec execution seat, 2026-09-16T20:47Z. PR #18528, head 3233eea6733eb621d76ed7386c558e290c987497, 17 files +494/-12, draft, --pair 18528 exit 0, needs:contract-review on both carriers.

⭐ The seat's error, owned

Mid-round the seat asked this dev to account for every file outside the claim's declared surface, in three buckets: generated, gate-required, or its own judgement call. The accounting came back 10 generated, 3 gate-required, 0 judgement calls — so the files the seat flagged as possible breaches were, with one exception, things the tree itself forces when a new schema is exported.

⇒ The claim's file-surface wording was narrower than what one new exported schema mechanically forces. That is the seat's drafting error, not a dev overreach, and it is recorded here rather than left as an unexplained flag on the PR. A surface that names src/ai/** and "the generated artifacts" without naming llms.txt, the alias-convention registry and the quick-reference counts is a surface that cannot be obeyed.

Bucket 1 (10), proven rather than asserted: every one reverted to base fb6b2c369e, re-produced by re-running its generator, and git diff HEAD came back empty — byte-identical. Generators named: gen:api-surface, gen:export-origins, gen:declaration-map, gen:schema, gen:strictness-ledger, gen:docs. ⭐ The restore used git restore --source=HEAD --staged --worktree rather than git checkout BASE -- path, because the latter stages what it writes — a detail that decides whether the tree is really clean afterwards.

Bucket 2 (3), each with the gate's own red quoted in the PR body: type-alias-convention.pin.test.ts (check:spec-parsed-alias — it is that gate's exemption registry), llms.txt (check:llms-txt, hand-kept by design — there is deliberately no generator, and it ships in the tarball), quick-reference.mdx (check:quick-reference-counts).

Bucket 3: none. quick-reference.mdx did carry a judgement call — an added table row — and the round dropped it on the seat's instruction, leaving only the count the gate forces (11 of 11 → 11 of 12, +1/-1), and corrected the PR body so it no longer describes a change the diff lacks. ⛔ Dropping the defence rather than defending the addition is the right move on a fence.

The reading gap, not papered over

The ruling lives on objectstack-ai/cloud#2172, and cloud is unreachable from this session. The round took the ruling text from the director seat's filing on this card, verbatim, and does not report the cloud issue as checked. objectui#7388 is reachable; it read it and all 10 comments.

Membership — measured, including what could not be sourced

member	source
structure / data / done	objectui plugin-chatbot/src/ChatbotEnhanced.tsx:163 (declared union) and mapMessages.ts:744 (d.phase === 'data' || d.phase === 'done' ? d.phase : 'structure'), read at objectui ff1d5ea8d1; structure is also that reader's coercion default
verify	objectui#7388 body verbatim, plus in-repo corroboration that service-ai-studio registers verify_build (platform-tool-names.ts)
could not source	the producer's own emission list — cloud unreachable. ⛔ Nothing was rounded out; the enum is exactly those four
explicitly rejected	designing — it belongs to the sibling data-blueprint-progress frame, and a test asserts it does not parse here

That is what "a closed enum is measured before it is written" was supposed to produce: four members with citations and one honest hole.

Evidence

New pin 16 tests exit 0; pnpm --filter @objectstack/spec test 484 files / 13789 tests; typecheck exit 0; repo-wide pnpm lint exit 0 run in full. 108 derived gate families, all 108 run, --ran reconciling "108 derived, 108 run, 0 NOT-MEASURED, 0 UNRUN". Five gates were initially prerequisite-not-met and the round built the prerequisites rather than reporting NOT MEASURED — all five then exit 0.

Two ablations from the committed state, each proving the mutation reached disk (occurrence counts both ways + a blob hash ≠ the HEAD blob) before reading any verdict: enum → z.string() = 3 failed (all three refusal cases, controls still green); looseObject → strictObject = 1 failed, exactly the FLOOR case. The looseObject choice is deliberate and now pinned: the panel fields objectui already ships ride the same frame, so a strict schema would refuse every frame in flight today.

The one non-zero gate is not this diff — and it is a TWIN

check:cross-package-test-inputs exit 1, naming packages/cli/... paths this diff never touches. Three-leg control on pristine origin/main 97233b90ca with zero of the round's changes: no dist/ → 0; only a built packages/spec/dist/ copied in → 1, byte-identical finding; dist/ removed → 0.

⭐ The #17931 round reached the same finding independently, in the same hour, on a different base (fb6b2c369e). The seat filed it then as #18526; this round's reproduction is corroboration on a second base, added there. ⛔ No second card — twin cards from parallel rounds are exactly what cross-reading reports is for.

The three "noted, not filed" calls, affirmed

• check:generated --fix does not flag declaration-map/ai.json stale when its companion export-origins/ai.json is reverted in the same stroke — a consistently-stale pair passes the aggregate. The round says it could not show a wrong verdict on a real workflow, so it recorded an observation rather than a defect. That is the right line: none of the three filing classes is met by "I could not make it wrong". ⚠️ Worth someone's attention if it ever produces a real stale landing — but a card whose evidence is "only visible because I reverted two files at once" would be a card nobody can act on.
• objectui's sibling extractBlueprintProgress collapse — already ruled on in objectui#7388, no producer ruling, no carrier. Correct to leave.
• contracts/ai-service.ts docblock still describes the frame without naming the enum — the claim permitted editing it; the round left it alone to keep this diff off a contracts/ path while PR docs(spec): scope the email-template locale-floor claims to a call that names a locale #18482 is in flight there. Conservative and fine; ⚠️ it means the channel's own docblock and the vocabulary that now exists are not yet cross-referenced. The seat carries that as a one-line follow-up, not a blocker.

State

CI on the final head was 11 success / 4 skipped / 17 in progress / 0 failure at 20:47:17Z and the round did not wait on it — convergence is the seat's, and it is being watched.

⚠️ Landing waits on the at-tier review, which cannot start. The contract-review tier is refusing at the account level (HTTP 429); references/contract-review.md L60 forbids a quota-exhaustion downgrade for review — 「豁免对象是派发,复核 ⛔ 不随派发档位免除」. Carriers stay hung on both sides, the PR stays draft and out of the queue. Details: #18503 (comment)

---

Generated by Claude Code