migrate meta: the stack-derived relevance proof that #20620's ruling names as the only exit for a notice is unbuilt — 17.7.0 still prints all 304 protocol-18 manual notices to an app none of them applies to #22072

Path: the road — upgrade: os migrate meta names what applies to this stack | 缺项 | P3

Triage: first grade, enhancement · priority:p3 · domain:spec · area:devpath · pm:queue. This builds the one exit #20620's ruling names: a structured, stack-derived relevance proof

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-07T13:04Z. ⛔ Not a claim, ⛔ not a dispatch.

Triage: lands in packages/spec/src/migrations/types.ts (SemanticMigration, :38) and packages/spec/src/migrations/chain.ts (where the list is assembled), plus the CLI flag in migrate meta ⇒ domain:spec; rationale: the entry type and the chain are the spec lane's. The flag rides the same PR.

• ⛔ Not a re-ruling of [finding][devx] os migrate meta --from 17 buries a project's real findings under 240 generic protocol-18 notices, and marks default-flip conversions as "Applied" when the right action is usually no edit #20620 (5888087153): no notice is dropped by prose matching. An entry leaves the default list only on a structured proof, evaluated over the loaded stack, that its surface is absent. Entries without a predicate stay listed.
• Why p3: the notices are correct, but none applies to the measured stack (304 of them on hotcrm, 17.7.0). The cost is authors learning to skip the command. No upgrade is broken.
• Scope:
  • The predicate field. An optional structured relevance predicate on SemanticMigration, the shape spec(migrations): a semantic entry names the D2 conversions whose applied edits it judges (SemanticMigration.conversionIds), so os migrate meta can pair them (the spec half of #20620) #20697 set for conversionIds: a named, testable question over the loaded stack. ⛔ No free text.
  • Evaluation in chain.ts.
  • The output prints the count of entries that left the default list. --all prints them anyway, so nothing becomes unreachable.
  • A first batch of predicates for the entries whose surface is a top-level stack key, with an enumeration pin. The PR body tables how many of the catalogue's entries carry a predicate, and hotcrm's listed count before and after.
• Clause-②: yes (a new optional field on a published type, and a new CLI flag). A contract review is owed at the PR.

Claim: PM loop round 1 · 2026-10-07T16:16Z
Session: session_01RPo7FUd6bSnAfkWMAKi848
Account: os-justin (the seat's linked user as GET /user answers it; the card's assignee from this act)
Branch: claude/issue-22072-migrate-meta-relevance-predicate
Worktree: objectstack-issue-22072
Domain: domain:spec
Seat: domain:spec#3 (seat post #18883)
File surface (at origin/main bafb58bb0; stop on breach and explain in the report):

• packages/spec/src/migrations/types.ts: an optional, structured relevance predicate on SemanticMigration (:38), in the shape spec(migrations): a semantic entry names the D2 conversions whose applied edits it judges (SemanticMigration.conversionIds), so os migrate meta can pair them (the spec half of #20620) #20697 set for conversionIds (:73). ⛔ No free text.
• packages/spec/src/migrations/chain.ts: evaluate the predicate over the loaded stack; an entry whose predicate is provably false leaves the default list; entries without a predicate stay listed.
• packages/spec/src/migrations/entries/semantic/** and packages/spec/src/migrations/registry.ts: the first batch of predicates, for entries whose surface is a top-level stack key, with an enumeration pin.
• Cross-lane (domain:cli): packages/cli/src/commands/migrate/meta.ts and its tests: print the count of entries that left the default list, and an --all flag that prints them anyway. Declared on the domain:cli seat post [PM seat] domain:cli — 🟢 os-warren · session_01RWZbGvPFcRKvUqASZtunCU #6024 in the same round.
• .changeset/22072-*.md (@objectstack/spec, @objectstack/cli).
  Container & model: M, mode:subagent, model: opus (--tier: no path-derived mandate; a clause-② widening, built at the default tier; the contract review is owed before enqueue from an isolated subagent at CONTRACT_REVIEW_TIER)
  Clause-②: yes
  Responsibility: n/a — not a defect card
  Thread-read: 6038524437
  Serial constraints cleared: packages/spec/src/migrations/registry.ts is in PR feat(spec)!: retire the flat string-list arm of manifest.permissions — the structured ADR-0025 block is the only form (#13458) #22094 (Phase 2 of #11333: retire the legacy string[] arm of manifest.permissions (major, standard retirement route) #13458, seat 2), PR feat(spec)!: the build doors refuse a builtin node config value its executor contract refuses, with its location #21974 (service-automation: a built-in node's config value its own contract refuses still registers, then fails every run — the built-in half of #21848's class #21898, seat 1's residual claim) and this seat's lint: promote relationship/master-detail-required from warning to error, scoped to controlled_by_parent — ruled for the v18 boundary (Direction 1 of #8772) #9139 branch (one new step-18 entry). Each adds or edits different entries: ordinary concurrency, not a single-claim path; whichever lands later merges main. No open PR touches types.ts, chain.ts or commands/migrate/meta.ts (open PRs' file lists read at this stamp).

Clause-②: yes per triage's grade 6038524437: "a new optional field on a published type, and a new CLI flag". The ruling it builds on is #20620's triage 5888087153: "an entry leaves the list only on a structured, stack-derived proof that its surface is absent. It is ⛔ never removed by prose-matching surface."

os-dev-report
{
"issue": 22072,
"status": "done",
"branch": "claude/issue-22072-migrate-meta-relevance-predicate",
"pr": "#22115",
"session": "session_01RPo7FUd6bSnAfkWMAKi848 — mode:subagent, the dispatching PM's harness-stamped id (container CLAUDE_CODE_REMOTE_SESSION_ID matches it); identity = the branch named by Claim 6041997447",
"premise_still_valid": true,
"summary": "SemanticMigration gains an optional relevance question, relevantWhen. It is a closed union whose only member this release is { kind: 'stack-declares', keys: [...] }, and keys are typed as StackDefinitionKey minus the carrier keys manifest/packages/plugins/devPlugins. applyMetaMigrations evaluates it in chain.ts with three values. The question is asked of the input stack and every hop checkpoint, and packages[].manifest bodies are read like the top level. These all answer unknown: any plugins/devPlugins entry, an unreadable value, and a non-plain stack. An entry that judges an applied conversion is always listed. Only absent moves an entry, from todos into the new absentTodos (on the chain result and on each hop). The CLI prints group ④ after the listed notices: a count line naming --all, plus a line scoping the proof to the loaded stack and not to stored rows. --all lists those entries in full, and --json always carries absentTodos. The first batch is 29 of 381 entries (5 of 77 in protocol 17, 24 of 304 in protocol 18), pinned by an enumeration test. On hotcrm c967803, --from 17 lists 304 before and 292 after (12 absent). That run reproduces the card's 304 manual notices and 12 applied edits exactly.",
"tests": "At head 1904d53, all real runs with the verdict lines quoted. (1) spec, the 57 test files importing MIGRATIONS_BY_MAJOR or applyMetaMigrations, the new src/migrations/semantic-relevance.test.ts included: 'Test Files 57 passed (57) / Tests 2284 passed (2284)', lock VERDICT command-exit 0. An earlier run of src/migrations/ alone was red on one pin (migrations.test.ts 'migrates the oldest supported major's shapes to canonical' asserted todos == the hop's whole semantic list); it was updated to the partition contract (todos plus absentTodos), and the 57-file run above is after it. (2) pnpm --filter @objectstack/spec typecheck: RC 0. (3) pnpm --filter @objectstack/cli exec vitest run --project unit --maxWorkers=2: 'Test Files 259 passed (259) / Tests 3800 passed (3800)'. (4) pnpm --filter @objectstack/cli typecheck: RC 0. The CLI reads result.absentTodos and hop.absentTodos, which only the rebuilt .d.ts declares, and that is the cross-package read-the-rebuilt-d.ts evidence. (5) The integration tier was owed because test/migrate-meta-engine-guidance.test.ts was edited: it now passes --all. Running it with test/migrate-meta-default-range.test.ts gave 'Test Files 2 passed (2) / Tests 10 passed | 1 skipped (11)'. test/migrate-meta.e2e.test.ts matches no local project (nightly): NOT MEASURED locally. (6) check:migration-registry: '✓ src/migrations/registry.ts is current (381 semantic, ...)'. check:generated first proved api-surface/ and export-origins/ stale, which were regenerated (+SemanticRelevance, +SemanticRelevanceKey, +StackDeclaresRelevance); the gate run then passed check:generated, check:api-surface, check:spec-changes and check:upgrade-guide (all exit 0). (7) ESLint, narrowed and proven: eslint --no-inline-config --format json on the 38 changed .ts files gave files=38 errors=0 warnings=0. The population is read from eslint.config.mjs files globs. Invariance: the config enables no type-aware linting (no parserOptions.project for any file), and its only disk reads at load are two baselines this diff does not touch, so no untouched file's verdict can move. (8) Measurements: this branch's CLI with --json, before = todos+absentTodos and after = todos. hotcrm c967803 (read-only public clone, spec resolved to this branch): from 17 gave 304 to 292, and from 16 gave 381 to 366. app-crm 304 to 291. app-todo 304 to 286. app-multi-package 304 to 280. app-showcase 304 to 304 (plugins make every answer unknown, so H2 is measured). A minimal one-object stack: from 17 gave 304 to 280, and from 16 gave 381 to 352. No ablation or reverse-mutation run: the new behaviour has no pre-existing implementation to ablate, and new-only fields are what the CLI typecheck reads.",
"mcp_calls": "1 — mcp__claude-code-remote__add_repo (objectstack-ai/hotcrm, access read; answered read_available, attached nothing). 0 MCP GitHub tools, read or write.",
"api_writes": "3 relay writes, all as objectstack-fleet[bot] through scripts/pm. (1) pr_create via scripts/pm/fleet-write/dispatch.mjs: POST /repos/objectstack-ai/objectstack/pulls (draft) opened #22115, and the tool read the body back byte-identical (13194/13194). (2) The assignee via scripts/pm/label-write.mjs --issue 22115 --assign os-justin: POST /repos//issues/22115/assignees, read back as assignees os-justin. (3) This os-dev-report comment via scripts/pm/post-stamped.mjs: POST /repos//issues/22072/comments. git push (branch probe plus 5 pushes) is not REST. No label written: none was named, and skip-changeset does not apply.",
"open_questions": [
{
"question": "For the contract review: applyMetaMigrations().todos, and the todos key of --json, now omit the entries proven absent; they move to the new absentTodos. Is that membership change of an existing output array a widening, as graded (Clause-②: yes, minor, no BREAKING line), or a narrowing?",
"options": [
"A: keep as graded — todos is the default list that the ruling lets an entry leave; absentTodos keeps every entry reachable, and the union is pinned equal to the old todos",
"B: keep todos complete, and add a per-entry absent marker plus a listed subset instead"
],
"recommendation": "A, because it is the output the triage scoped. The union-equals-old pin plus --json always carrying absentTodos means no consumer loses an entry, only the default reading narrows to what the stack may owe. Non-blocking: shipped as A."
}
],
"out_of_scope_findings": [
"class: a · reach: public door, os migrate meta --from 18 --to 18 --out FILE (a range with no step) on a one-object stack exits 0, writes no snapshot and prints no line about it, while the same command with --json does write FILE. The human branch for an empty range (printMigrationReport's early return) never reaches the --out write. This is pre-existing: that branch is unchanged by this PR, since an empty range has no absentTodos. · evidence: measured in this run with the branch CLI; 'ls snap-empty.json: No such file', and the --json variant produced snap-json.json · dedupe words: migrate meta --out not written empty range · nothing-to-migrate branch skips --out · printEmptyRangeAnswer snapshot",
"carrier: none (承接者:无) · noted, not filed — content/docs/upgrading.mdx 'Useful flags' table does not list --all. It is outside the claimed file surface, so it is recorded in the PR's Acceptance notes only.",
"carrier: none (承接者:无) · noted, not filed — the D3 semantic catalogue reaches an operator only through the authored-source run, while --stored reports D2 conversion todos per row and never D3 entries. A source-scoped relevance proof therefore says nothing about Studio, metadata-API or AI-authored rows. The printed scope line states this boundary. Observation, not a defect class.",
"carrier: none (承接者:无) · noted, not filed, not reproduced — the chain's conversions walk top-level collections only, so an assembled packages[] stack (examples/app-multi-package's config is composeStacks with manifest 'preserve') would get no mechanical rewrite inside package bodies. I did not build a repro, so it stays an observation in the Acceptance notes."
],
"gates": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 1904d53 derived 93 families. 92 exited 0. 1 is NOT MEASURED: pnpm check:dual-build-cjs-loads exit 3 PREREQUISITE NOT MET, because it needs every workspace package built (studio, client-react, connector-slack, ... have no dist) and this tree built the CLI and showcase closures only. 17 families overlapped an in-flight rebuild of spec/dist (the showcase closure build) and were re-run after it, all exit 0. Of those, i18n-coverage, i18n-walk-parity and lean-entry-closure had first read a vanished dist (exit 3/3/1). Reconciled with --ran using ' :: exit N' records: '93 derived famil(ies) accounted for — 92 run, 1 NOT-MEASURED (1 DERIVED from a recorded exit 3)'. Every exit code was captured before any pipe (cmd > log 2>&1; rc=$?).",
"files_changed": "41 files, +867 / -56: packages/spec/src/migrations/types.ts +91 -1; chain.ts +156 -9; index.ts +3; registry.ts +29 (generated); entries/semantic/** 29 files +29; semantic-relevance.test.ts +294 (new); migrations.test.ts +6 -3; packages/spec/api-surface/migrations.json +3 and export-origins/migrations.json +3 (generated); packages/cli/src/commands/migrate/meta.ts +99 -21; meta.report-order.test.ts +126 -14; packages/cli/test/migrate-meta-engine-guidance.test.ts +10 -8; .changeset/22072-migrate-meta-relevance-predicate.md +18 (spec minor, cli minor, Clause-②: yes)",
"deviations": [
"Commit trailers use the model-free pair AGENTS.md prescribes (Co-authored-by: Claude, Claude-Session session URL). The harness reminder asked for a model-named Co-Authored-By, which the pre-push hook refuses. No card trailer.",
"hotcrm's before/after was measured on its read-only clone at c967803, loaded by this branch's CLI with @objectstack/spec resolved to this branch, not on the 17.7.0 tarball. The baseline reproduced exactly (304 manual, 12 applied).",
"The branch was not merged with main: it is 8 commits behind. A local merge-tree with origin/main is clean, registry.ts included (main added one step-18 entry with no question). The PM's concurrency note leaves the merge to whichever PR lands later.",
"There was no separate reverse-mutation run for the cross-package type change. The evidence is that the CLI typecheck passes only by reading absentTodos and relevantWhen, which exist only in the rebuilt .d.ts.",
"The worktree was cleaned: node_modules removed, then git worktree remove (no --force), exit 0. The scratchpad hotcrm clone was removed."
]
}

Generated by Claude Code

✅ ACCEPT: PR #22115 at 1904d53507. ⛔ Not enqueued yet: the contract review is owed (Clause-②: yes) and CI is converging

domain:spec seat 3 (#18883) · os-justin · session session_01RPo7FUd6bSnAfkWMAKi848 · 2026-10-07T18:29Z · holder of claim 6041997447; the review of record for the report 6044219249.

Checklist (read on GitHub and the PR's own diff, not from the report):

• Form: draft, base main, first line Fixes #22072, no other closing keyword; Clause-②: yes starts a line. PR assignee os-justin.
• Scope: 41 files, +867 / −56, no governed path, all inside the claim's surface:
  • migrations/types.ts, chain.ts, index.ts;
  • 29 one-line relevantWhen additions under entries/semantic/**, with the generated registry.ts, api-surface and export-origins;
  • the cross-lane commands/migrate/meta.ts (declared on [PM seat] domain:cli — 🟢 os-warren · session_01RWZbGvPFcRKvUqASZtunCU #6024, 6042007377) and its tests;
  • one changeset.
• The ruling's shape holds: relevantWhen is a closed union with one member, { kind: 'stack-declares', keys }. There is no free text and no matching on surface. An entry leaves the default list only on an absent answer; plugins, unreadable values and non-plain stacks answer unknown and stay listed ([finding][devx] os migrate meta --from 17 buries a project's real findings under 240 generic protocol-18 notices, and marks default-flip conversions as "Applied" when the right action is usually no edit #20620 5888087153). absentTodos keeps every entry reachable, and --all prints them.
• First batch: 29 of 381 entries, pinned by an enumeration test. Measured (report): hotcrm --from 17 lists 304 → 292, app-crm 304 → 291, app-showcase 304 → 304 (its plugins make every answer unknown, which is the conservative case working).
• Changeset: @objectstack/spec and @objectstack/cli minor, Clause-②: yes. Read sentence by sentence against the diff: the summary line, the four bullets (the field and its exported types; applyMetaMigrations's partition and the three unknown cases; the 29-entry batch and its key list; the CLI's count line, scope line, --all, --json and --step behaviour). Each matches the code.

Open question, for the contract review (not answered by this seat): applyMetaMigrations().todos now omits the entries proven absent, which move to absentTodos. The union equals the old todos (pinned). Is that membership change of an existing output array the widening the claim graded, or a narrowing a programmatic consumer could feel? The at-tier review rules on it before enqueue.

Out of scope:

• The --out file not written on a range with no step: filed finding(cli): os migrate meta --out FILE over a range with no step exits 0 and writes no snapshot, while the same run with --json writes FILE #22116 (class a, measured at the CLI door; pre-existing; domain:cli by its landing site, for triage to grade).
• content/docs/upgrading.mdx's flag table does not list --all: Acceptance notes (carrier: none).
• --stored never reports D3 entries, so the relevance proof covers authored source only: Acceptance notes. The CLI's printed scope line states that boundary.
• packages[] bodies get no mechanical conversion pass: Acceptance notes, not reproduced.

Landing, held until:

• a ## Contract review PASS at CONTRACT_REVIEW_TIER is on record for the landing head (dispatched to an isolated at-tier reviewer);
• every check is green on that head (17 of 32 were in_progress at this stamp).

The branch is 8 commits behind main. The report shows a clean local merge-tree, and the queue rebuilds on current main.

🔁 REWORK (patch round 1): PR #22115 at 1904d53507, after the contract review's FAIL 6044604409. Same claim, same branch, the same dev is re-dispatched

domain:spec seat 3 (#18883) · os-justin · session session_01RPo7FUd6bSnAfkWMAKi848 · 2026-10-07T18:51Z · holder of claim 6041997447.

The at-tier review found the types, the evaluation, the partition mechanics, the CLI output and the pins right as read, and two things wrong. The patch round carries exactly these:

1. Four first-batch entries come off. Remove relevantWhen from cel-predicate-one-value-comparand-refused, cel-predicate-variable-root-comparand-refused, rls-predicate-array-comparand-refused and rls-predicate-stored-list-ordering-refused.
   • Each one's surface names a code door (a direct matchesFilterCondition / compiler caller), so "the stack declares no such key" does not bound it.
   • Then regenerate registry.ts, set the enumeration pin to 25, correct the PR body's table and the changeset's count, and re-run the measurements.
2. The published todos guarantee stays whole: the review's disposition B, the seat's choice.
   • applyMetaMigrations().todos, MigrationHopResult.todos and --json's todos keep every semantic entry, as their TSDoc promised ("Every semantic TODO across all hops").
   • The proven-absent entries are named in absentTodos / hops[].absentTodos as a subset of todos.
   • The CLI's default list prints todos minus absentTodos; --all prints them.
   • Why B over A′:
     • it narrows no published guarantee, so Clause-②: yes, minor with no BREAKING line stays exactly right;
     • the published upgrade skill (skills/objectstack-upgrade/SKILL.md, Tier H) loops over --json's todos, and B leaves its reading unchanged, with no governed edit.
   • The changeset sentence names absentTodos as a required member added to the two result interfaces.
3. tiers counts as a plugin carrier (review ③13), so its presence answers unknown, like plugins / devPlugins. This is in proportion: one more key in the same conservative check, pinned.
4. Ride-along (review ① item 7): one row for --all in content/docs/upgrading.mdx's "Useful flags" table.

After the push: a fresh ## Contract review on the new head, then landing as before. needs:contract-review stays on the PR (FAIL does not remove it).

os-dev-report
{
"issue": 22072,
"status": "done",
"patch_round": "1 — answers the contract review FAIL 6044604409 with the seat's REWORK 6044626135 (dispositions 1-4 exactly)",
"branch": "claude/issue-22072-migrate-meta-relevance-predicate",
"pr": "#22115",
"head": "5104a5d74 (on merge de8ccbc of origin/main a543e24 into 1904d53; pushed fast-forward, no force)",
"session": "session_01RPo7FUd6bSnAfkWMAKi848 — mode:subagent, the dispatching PM's harness-stamped id; identity = the branch named by Claim 6041997447 (still the newest Claim on the card)",
"premise_still_valid": true,
"summary": "I made the four REWORK changes and nothing else. (1) relevantWhen is removed from cel-predicate-one-value-comparand-refused, cel-predicate-variable-root-comparand-refused, rls-predicate-array-comparand-refused and rls-predicate-stored-list-ordering-refused. Each file is byte-identical to its pre-PR blob at bafb58b. registry.ts is regenerated, and the enumeration pin is 25 (17: 5, 18: 20). (2) Disposition B. applyMetaMigrations().todos, MigrationHopResult.todos and --json todos keep every semantic entry again. migrations.test.ts's original equality pin is restored byte-for-byte, with a subset pin added. absentTodos and hops[].absentTodos now NAME a subset of todos (the same objects, chain order). The CLI lists todos minus absentTodos, ③'s header counts that difference, and ④'s count line is unchanged. The TSDoc on both result interfaces and the changeset say this, and name absentTodos as a required member added to both interfaces. The grade stays Clause-②: yes, minor, no BREAKING line. (3) A declared non-empty tiers answers unknown like plugins/devPlugins. tiers is also excluded from SemanticRelevanceKey. Both are pinned. (4) The --all row is added to content/docs/upgrading.mdx 'Useful flags'. I merged origin/main first via os-regen-merge.sh (clean). main added two step-18 entries, so the catalogue is now 383 (77 + 306).",
"tests": "At head 5104a5d; every exit code was captured before any pipe. Build: turbo build of the cli closure + cli + showcase closure gave RC 0. check:generated: 'All 15 generated artifacts are up to date', so no artifact needed regenerating. Spec: the 58 test files importing MIGRATIONS_BY_MAJOR or applyMetaMigrations (semantic-relevance.test.ts and migrations.test.ts included) gave 'Test Files 58 passed (58) / Tests 2331 passed (2331)'. spec typecheck RC 0. cli typecheck RC 0. The cli unit tier gave 'Test Files 1 failed | 260 passed (261) / Tests 2 failed | 3841 passed (3843)'. The one failing file is test/hook-timeout-override-refusal.test.ts: two CONTROL cases 'Test timed out in 5000ms' while the box ran 117 gate families. That file is untouched by this diff and by the merge, and its rerun alone under the lock gave 'Test Files 1 passed (1) / Tests 4 passed (4)', VERDICT command-exit 0. So I read it as load-induced, and it is not a regression. cli integration (test/migrate-meta-engine-guidance.test.ts + test/migrate-meta-default-range.test.ts) gave 'Test Files 2 passed (2) / Tests 10 passed | 1 skipped (11)'. ESLint narrowed: eslint --no-inline-config --format json on the 34 changed TS files vs merge-base a543e24 gave files=34 errors=0 warnings=0. The config has no type-aware linting, and its only load-time disk reads are two untouched baselines, so no untouched file's verdict can move. Measurements: this branch's CLI with --json; before = todos (whole), after = todos - absentTodos, and subset=true on every row. hotcrm c967803 (public, re-cloned read-only, spec resolved to this branch): from 17 gave 306 to 294 (12), from 16 gave 383 to 368 (15), applied 12. app-crm 306 to 293 (13). app-todo 306 to 292 (14). app-multi-package 306 to 286 (20). app-showcase 306 to 306 (0; it lists runtime plugins). A minimal one-object stack: from 17 gave 306 to 286 (20), from 16 gave 383 to 358 (25). hotcrm's --step hop line reads '12 mechanical, 294 manual, 12 not listed (surface absent)'.",
"mcp_calls": "0 this round (0 MCP GitHub tools). Round 0's single read-only add_repo is not repeated; hotcrm was re-cloned through the anonymous git lane it had answered.",
"api_writes": "1 this round: this os-dev-report comment, via scripts/pm/post-stamped.mjs, as POST /repos//issues/22072/comments through the fleet relay. Nothing else is written: the PR body is left to the seat (see pr_body_changes), and no label or assignee is touched. git push of 2 commits (the merge and the patch) is not REST.",
"pr_body_changes": "For the seat to apply to PR #22115's body. (a) The batch table: delete the 'permissions / sharingRules' row. Reduce the 'permissions' row to 1 entry, permission-restore-purge-bits-retired (rls-predicate-array-comparand-refused and rls-predicate-stored-list-ordering-refused come off). Make the total row '5 of 77 · 20 of 306 · 25 of 383'. (b) In the left-out list under rule 2, add the four as 'their surface also names a code door: a direct matchesFilterCondition / compiler caller'. (c) The evaluation table: the plugins row becomes 'any entry in plugins / devPlugins / tiers'. (d) The evaluation and output bullets: 'todos keeps every entry; absentTodos names the proven-absent subset (same objects)'. The CLI 'lists todos minus absentTodos'. Keep 'a run whose only notices are proven absent still counts them and writes --out'. Drop the clause saying the 'Nothing to migrate' condition changed: with todos whole, that condition is back to the original byte-for-byte, and such a run reaches ④ through the ordinary path. (e) The measured table, replaced by the subset readings above (306 / 383 bases). Note that main's two new step-18 entries move the hotcrm base from 304 to 306, and the 12 left are unchanged. (f) 'For the contract review': the todos-membership bullet becomes 'disposition B: todos and --json todos unchanged; absentTodos is a new required member of MigrationChainResult and MigrationHopResult'. (g) Verification and gates: the numbers above, ESLint at 34 files, and gates at 117 families. (h) Acceptance notes: drop the docs-table note, which is now fixed, and drop the upgrade-skill note, whose reading is unchanged under B.",
"open_questions": [],
"out_of_scope_findings": [
"carrier: the seat — no new finding this round. Round 0's class-a finding (--out not written on an empty range) is already filed by the seat as #22116; it is not addressed here, and the branch that holds it is unchanged."
],
"gates": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 5104a5d derived 117 families (93 last round, plus the docs families the upgrading.mdx row brings in). 116 run exit 0. 1 NOT MEASURED: pnpm check:dual-build-cjs-loads exit 3, PREREQUISITE NOT MET, because it needs every workspace package built (studio etc. have no dist). Two families needed one rerun on a prerequisite, and both are counted at their rerun's code. node scripts/check-comment-mask-corpus.mjs first exited 1 on ENOENT for a transient objectstack.config.bundled_*.mjs that my own concurrent measurement run created in examples/app-showcase; rerun it gave exit 0, '8341 files, 0 disagree'. pnpm --filter @objectstack/spec run check:skill-examples first exited 3 because client-react was not built; after pnpm --filter @objectstack/client-react build under the lock it gave exit 0, '262 prose examples type-check'. --ran with ' :: exit N' records: '117 derived famil(ies) accounted for — 116 run, 1 NOT-MEASURED (1 DERIVED from a recorded exit 3)'.",
"files_changed": "Patch commit 5104a5d: 13 files, +204 / -141. PR vs merge-base a543e24: 38 files, +937 / -63. That is .changeset/22072-migrate-meta-relevance-predicate.md +25; content/docs/upgrading.mdx +1; packages/cli/src/commands/migrate/meta.ts +117 -25; meta.report-order.test.ts +151 -20; packages/cli/test/migrate-meta-engine-guidance.test.ts +10 -8; packages/spec/src/migrations/types.ts +100 -1; chain.ts +156 -9; index.ts +3; migrations.test.ts +4 -0 (the original pin restored, plus subset lines only); semantic-relevance.test.ts +314 (new); registry.ts +25 (generated); entries/semantic 25 files +25; api-surface/migrations.json +3; export-origins/migrations.json +3.",
"deviations": [
"The unit tier has one load-timed-out file, hook-timeout-override-refusal.test.ts, at 5000ms in the full run, and it passes alone (4/4). I recorded it as such, not as green for the full run.",
"hotcrm was measured on a read-only public clone at c967803 against this branch's spec, as in round 0. Its base is 306 now because main added two step-18 entries since the card's 17.7.0 reading.",
"I did not edit the PR body (per the REWORK); the requested body changes are in pr_body_changes.",
"The worktree was cleaned again: node_modules removed, then git worktree remove without --force, exit 0. The scratchpad hotcrm clone was removed."
]
}

Generated by Claude Code

✅ ACCEPT (patch round 1): PR #22115 at 5104a5d74. ⛔ Not enqueued: a fresh contract review is owed on this head

domain:spec seat 3 (#18883) · os-justin · session session_01RPo7FUd6bSnAfkWMAKi848 · 2026-10-07T19:59Z · holder of claim 6041997447; the review of record for the patch-round report 6045718354.

Each REWORK item (6044626135) checked on the PR's own tree, not from the report:

1. The four code-door entries (cel-predicate-one-value-comparand-refused, cel-predicate-variable-root-comparand-refused, rls-predicate-array-comparand-refused, rls-predicate-stored-list-ordering-refused) are blob-equal to their pre-PR blobs at bafb58bb0. 25 entries carry relevantWhen.
2. Disposition B: todos is whole again. migrations.test.ts keeps its original equality pin and adds a subset pin (absentTodos ⊆ todos, each carrying relevantWhen). chain.ts pushes the same objects into absentTodos.
3. tiers is a carrier: chain.ts iterates ['plugins', 'devPlugins', 'tiers'].
4. Docs: content/docs/upgrading.mdx gains the --all row. Read against the CLI: it lists them in full, --json keeps them in todos and names them in absentTodos.

The PR body was rewritten by this seat from the dev's pr_body_changes (a)–(h): the batch table at 25 of 383, the left-out list, the evaluation rows, the measured table on the 306 / 383 bases, the contract-review bullet, verification at this head, and the acceptance notes. Read back byte-identical.

One recorded gap: in the dev's full local run, packages/cli/test/hook-timeout-override-refusal.test.ts timed out twice under box load, and it passes alone (4/4). That file is untouched by the diff and by the merge, and CI's Test Core is the reading that counts.

CI on this head at this stamp: 33 success, 2 skipped, 0 failed. Landing waits for a ## Contract review PASS on 5104a5d74 (dispatched now) and every check green.

Landed: PR #22115 → cdeabec848. The card closes completed

domain:spec seat 3 (#18883) · os-justin · session session_01RPo7FUd6bSnAfkWMAKi848 · 2026-10-07T23:18Z · holder of claim 6041997447.

• Landed: PR feat(spec,cli): a structured relevance question takes a semantic notice off os migrate meta's default list (counted; --all lists it) #22115 merged through the merge queue at 2026-10-07T23:17Z as cdeabec848. It has one parent, 54ace18c6, and is an ancestor of origin/main.
• Content check: all 39 files the squash changed against 54ace18c6 are blob-equal to the reviewed head 16cb55b8d0 (contract review PASS 6048283089 on that head; the earlier PASS 6045970836 covered 5104a5d74). Between the two records the branch merged main twice and carried one changeset line (1d6efd229) and one test line (16cb55b8d).
• What now holds:
  • A semantic entry may carry relevantWhen: { kind: 'stack-declares', keys: […] }, a closed question that os migrate meta answers from the stack.
  • 25 of 391 entries carry one (5 at step 17, 20 at step 18).
  • A notice proven absent leaves the default list and is counted. --all lists it, and --json names it in absentTodos as a subset of an unchanged todos.
  • It ships minor for @objectstack/spec and @objectstack/cli, with Clause-②: yes and no BREAKING line.

A correction to the seat's own reading. On the CI failure of 1d6efd229 (Type Check · workspace), the seat's merge-round order guessed that #22121's new migrate-meta-out-snapshot.test.ts lacked absentTodos. The dev measured TS2741 on all: this PR's required MigrationReport.all, a CLI-internal type. The fix was the one line all: false,. The order's premise was wrong, and the measured cause is the one that landed.

Acceptance notes (nothing filed):

• fix(cli): os migrate meta --out writes its snapshot on a run with nothing to migrate #22121's synthetic early-return arm in migrate-meta-out-snapshot.test.ts keeps a non-empty absentTodos beside todos: [], a state outside this PR's subset contract. Its assertions hold either way, so it is left as written.
• Standing review observations, non-blocking: printEmptyRangeAnswer's wider-range hint counts listed notices only (unreachable at 25 of 391 questioned). --preset is not addressable by a stack-derived question.
• Dev report out-of-scope items: --stored never reports the D3 catalogue (an observation; the scope line bounds the proof). packages[] bodies get no mechanical conversion pass (pre-existing and not reproduced; filed once someone reproduces it).

This act removes pm:dispatched from the closed card. domain:spec, area:devpath and the type label stay.