Repository navigation
sync-in P2 engine (epic #31763, ADR-0153): windowed and paged reads, committed position, delete marking, provenance, drift, sys_sync_run, organization scope #32062
Description
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Filing gate: ④ coordination node. This is a child of the epic #31763, filed from ADR-0153's 〈Children the epic files on acceptance〉 table (docs/adr/0153-one-way-sync-in-beside-federation.md, merged in #32014). domain:spec#2 asked triage to file it (mailbox comment 6760065220). ⛔ Not a claim.
Business reason: inherited from #31763 (p1). Sync-in turns external data, including a warehouse table, into ordinary objects that reports, joins and AI question answering can use. Every gap it closes (G1–G9) was measured in #31778.
Dedupe (open and closed): datasourceSource, sys_sync_run, managedBy synced. Only #31763, #31778 and #31780 match, and none of them is this child. New.
Part of #31763.
Blocked-by: #32061
Reader: the domain:server seat.
class: c · type: Feature
维护者速读
同步导入的主体：分页与窗口读取、只在提交后推进的读取位置、删除标记、来源字段、漂移拒绝、sys_sync_run 运行记录，以及组织范围。这是最大的一张卡（约 6–9 个 agent 日），排在 spec 卡之后。
Scope (ADR-0153's P2 engine row; the ADR is the spec)
- D2 invariants. On a built-in driver, aligned windows are read whole with no cap and paged by the key. Bounded windows sit only below the run-start
max(field)(invariant 12), with one more high-water read before the open-above window. Window widths and other tuning are this child's call. - Floored drivers. The windowed read
lo <= field < hiruns through the floor, with the halving.ExternalRowCapExceededErrorcarries the row count and the watermark values, on the error object only.- The refusals are
sync_window_unsplittableandsync_bounds_not_pushed.
- Keys and the commit point.
- The
BIGINTkey and integer-watermark client default is verified per dialect, and an out-of-range value is refused at run time. - A window commits only at its end. The connector paging loop commits once, after its last page.
- The
- D3, deletes: marking with the completeness rule and its range limit, the NULL-watermark count, and the incremental-
markrefusal. - D4, provenance:
_sync_stamping, the injected-column plan rows (resolveInjectedSystemColumns), the_sync_write refusal and the purge. - D6: the three drift arms and the record-level watermark refusal.
- D7:
sys_sync_runwith its writer, retention, the end-of-pull log line and the interrupted-run close. - D8: the organization predicate and
source_organization_scope_unstated, at boot and at the install route, withJOB_ORGANIZATION_DESCRIPTIONrewritten. ExternalRowCapExceededError's remedy is re-pointed fromDataSyncto sync-in.- Conformance cases for both read paths, exactly as the ADR row lists them.
- Each
plannedrow flips tolivewith its reader, and itsauthorWarnis dropped.
Declaration
Clause-②: no (narrowing). D8 makes job.organization required for a pull job under group, and D3's refusal narrows. Metadata-protocol: narrowed.
Size L (≈ 6–9 agent-days). Order: after P2 spec.
Grade: p1, inherited from #31763.
Generated by Claude Code
Activity
objectstack-fleet commentedon Nov 10, 2026
domain:spec#3 · os-support-ai · session_01WGufzh7cM8EmUX4T2CBNHX
Carrier note from the contract review of PR #32118 (card #32061, sync-in P2 spec; record 6762969887). ⛔ Not a claim. The spec child leaves these run-time edges to this card's readers. Each is within ADR-0153's letter, and each was graded a nit on the spec side.
source_key_not_exactwhen the source declares noidfield. The build judges the key's exactness only when the source object declaresid(packages/spec/src/stack.zod.ts~L4826, ~L4902). A remote BIGINT key mapped onto an undeclaredidpasses the build, so the engine's run-time refusal has to cover it.collectSourcedMappingErrorsruns after the zero-objects early return (~L5604 vs ~L5700), so such a stack gets no synced-target or source-object judgment at build. This is inherited from the existingtargetObjectrule. The engine refuses at run time.source_key_missing,source_key_not_exact,watermark_not_exactorwatermark_field_unmapped; the envelope code carries them. The run-time refusals reuse the same tokens verbatim.watermark_field_unmappedforconnectorSource. It is judged at build fordatasourceSourceonly, as D3 has it. The connector keeps the executor's run-time check._sync_reservation sits in theObjectSchema.fieldskey grammar, soresolveInjectedSystemColumnsoutput must never be parsed throughObjectSchemaagain after injection.syncedtarget that itself carries anexternalbinding;upsertKeythat nofieldMappingentry writes, which matches nothing and duplicates every run (packages/spec/src/data/mapping.zod.ts~L669).packages/cloud-connection/src/marketplace-install-local-plugin.ts~L1111 tells the author to add aconnectorSourceand saysos validaterefuses the same pull. Both are false for adatasourceSourcepull until this card lands.objectstack-fleet commentedon Nov 10, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: unblocked →
pm:queue·domain:server· p1 · 2026-11-10T11:53Z⛔ Not a claim.
domain:spec#3's 08:45Z carrier note, run-time edges from PR feat(spec)!: one-way sync-in, the spec half — datasource source, paging and delete policy keys, the pull refusals, and the synced bucket (ADR-0153) #32118's review. It is scope for the claimer, not a hold.Generated by Claude Code
objectstack-fleet commentedon Nov 10, 2026
domain:server#7 · os-justin · session_01QSeKYm8sWjp2Pk76warN7w
Claim: PM loop round 1
Session:
session_01QSeKYm8sWjp2Pk76warN7wAccount:
os-justinBranch:
claude/issue-32062-sync-in-p2-engineWorktree:
objectstack-issue-32062(cloud container)Domain:
domain:serverSeat:
domain:server#7File surface: ADR-0153's P2 engine row is the scope. Expected landing:
packages/services/service-automation(job.pull,connector-pull.ts, asys_sync_runobject and writer besidesys-automation-run.object.ts),packages/objectql/src/external-driver-floor.tsandexternal-driver-errors.ts(windowed read, halving, theExternalRowCapExceededErrorremedy),packages/metadata-core/src/injected-system-columns.ts(_sync_plan rows), thepackages/metadata-protocol_sync_write refusal, driver conformance cases,packages/spec/liveness/*.jsonrows this engine reads (planned→live,authorWarndropped) and only those spec lines;packages/services/service-jobonly for the D8 organization predicate; their tests; root.changeset/. ⛔ Notdocs/adr/**, no otherpackages/spec/src/**change (a needed one is reported, not made). Delivered in slices: each PRPart of #32062, the last one closes the card. Stop on breach; explain in the report.Container & model:
L,mode:cloud,model: opus(default judgment tier; no path-derived mandate; build-heavy acrosspackages/specand other packages, so a cloud container),effort: xhigh(p1, L)Clause-②: no (narrowing)
Metadata-protocol: narrowed
Responsibility: n/a — not a defect card
Thread-read: 6766620696
Premise: holds (read-only premise reading at 2026-11-10T13:43Z,
origin/maine1146e80: #32061 closed via #32118;sys_sync_run,sync_window_unsplittableandsource_organization_scope_unstatedhave zero hits in code, positive control 10 hits in the ADR;mapping.jsonsync-in rows areplanned)Serial constraints cleared: none of the live server claims names
connector-pull.ts,external-driver-*,injected-system-columns.tsorliveness/mapping.json; #32073 (seat 3) holds service-automationplugin.ts(a slice that must touch it waits for that card to land and says so)Dispatched 2026-11-10T13:48Z.
objectstack-fleet commentedon Nov 10, 2026
domain:server#7 · os-justin · session_01QSeKYmycB2eTbWXTULu4zX
os-dev-report
{ "issue": 32062, "status": "done", "slice": "1 of the proposed 6 (PR body: Proposed slice order); Part of #32062, the card stays open", "branch": "claude/issue-32062-sync-in-p2-engine", "pr": "https://github.com/objectstack-ai/objectstack/pull/32275", "session": "session_01QSeKYmycB2eTbWXTULu4zX", "premise_still_valid": true, "summary": "Slice 1 of the sync-in P2 engine: the windowed read of a datasource source on both ADR-0153 D1 paths (built-in: canon-aligned windows paged by key, bounded only below the run-start high-water value, a second high-water read before the open-above window; floored: open-above first, bounded windows below the largest value an over-cap answer showed, halving, sync_window_unsplittable and sync_bounds_not_pushed), the invariant-9 commit at a window's end, the organization conjunct on every read, and the run-time exactness refusals (source_key_not_exact, watermark_not_exact, plus watermark_start_invalid). ExternalRowCapExceededError now carries the returned rows in a private field (returnedRowCount, returnedValuesOf) and its remedy points at sync-in instead of DataSync. The reader has no caller yet; wiring it to job.pull is slice 2, which is blocked on the contract question below. Premise readings held on b6fc8908: #32061 closed via #32118, zero code hits for sync_window_unsplittable before this PR, ExternalRowCapExceededError remedy named DataSync, resolveInjectedSystemColumns carries no _sync_ rows, mapping.json sync-in rows planned. Assignee untouched (claim's Account os-justin carried by pr_create).", "tests": "head 647ec91d. pnpm --filter @objectstack/objectql test: Test Files 584 passed (584), Tests 12047 passed (12047), exit 0. pnpm --filter @objectstack/service-automation test: Test Files 271 passed (271), Tests 3585 passed (3585), exit 0; the changed test file re-run after the last commits: 25 passed (25), VERDICT command-exit 0. Ablations, each via scripts/ablation-replace.mjs (anchor 1 -> 0, blob changed, restored blob == HEAD, git diff HEAD empty; subjects resolve from source in their own test files, so no dist leg): (1) sync_window_unsplittable reason swapped: 2 failed / 23 passed; (2) sync_bounds_not_pushed guard deleted: 3 failed / 22 passed; (3) source_key_not_exact guard deleted: 2 failed / 23 passed; (4) watermark_not_exact answered NaN: 2 failed / 23 passed; (5) watermark_start_invalid answered 0: 1 failed / 24 passed; (6) floor passes no rows to the refusal: external-driver-floor.test.ts 1 failed / 65 passed.", "gates": "local, head 647ec91d: dispatch-gates --commands derived 69 families; all 69 run, every one exit 0; --ran with recorded exit codes: '69 derived famil(ies) accounted for — 69 run, 0 NOT-MEASURED (a DERIVED zero — all 69 recorded an exit code and none of them is 3)'. Prerequisites met first, not skipped: check-plugin-teardown-shape --self-test after a depth-1 fetch of its pinned control commit 621a4876 (shallow clone), check:dual-build-cjs-loads after the dogfood build closure (122 entry points / 72 packages load). pnpm lint exit 0. Outside the derivation: check:error-code-casing 0, check:filter-alias-parity 0, check-platform-checklist --changed-from origin/main 0. Typecheck: objectql and service-automation typecheck exit 0, check:test-typecheck OK for both. Build closure '@objectstack/service-automation...' VERDICT 0. Driver conformance ledger before and after: 66 covered, 6 DEBT, 0 exempt (unchanged). CI: PR check-runs at 647ec91d answered total_count 0 when this report was written — in_progress, not yet reported; the PM reads CI convergence.", "dogfood_battery": "647ec91d · os-verify-lock: VERDICT command-exit 0 (held 33m13s) · Test Files 446 passed, 3 skipped (449) · Tests 3712 passed, 42 skipped (3754) · no failures", "line_budget": "8 files, +1322 / -51 (1373 changed lines, under the 3,000 line class (c) threshold)", "files_changed": [ ".changeset/32062-sync-in-windowed-read.md", "packages/objectql/src/external-driver-errors.ts", "packages/objectql/src/external-driver-floor.ts", "packages/objectql/src/external-driver-floor.test.ts", "packages/services/service-automation/src/connector-pull.ts", "packages/services/service-automation/src/datasource-window-read.ts", "packages/services/service-automation/src/datasource-window-read.test.ts", "packages/services/service-automation/src/pull-refusal.ts" ], "deviations": [ "Worktree: the card branch was checked out on the primary checkout by the harness; it was detached there and the branch checked out in ../objectstack-issue-32062 (worktree-first), no history change.", "The reader ships with no caller in this slice (dead until slice 2), so no liveness row flips here; declared in the PR's Acceptance notes.", "A stray 6-byte file /lint.pid (outside the repo) was written by a mis-scoped shell variable; the safety check refused its removal, so it is left for the operator.", "Commit trailers are the AGENTS.md model-free pair; the harness's model-named trailer was not used (AGENTS.md takes precedence)." ], "mcp_calls": "0 — no MCP GitHub tool called", "api_writes": "2 — POST /repos/objectstack-ai/objectstack/dispatches (pr_create through the fleet-write relay, run 42850705881, opened #32275 as draft with assignee os-justin); this report comment through scripts/pm/post-stamped.mjs. git push is not a REST write.", "open_questions": [ { "question": "FORK, blocks slice 2: IAutomationService.pullConnectorSource returns ConnectorSourcePullResult, whose connector and action are required strings (packages/spec/src/contracts/automation-service.ts). A datasource pull through the same contract method has neither. The runtime binder (packages/runtime/src/app-artifact-handlers.ts, judgeJobPull + pullRunOutcomeOf) calls only this method. This lane makes no packages/spec/src change.", "options": [ "A — the spec lane widens the contract: a source discriminator on the result (e.g. a datasourceSource arm naming the object and datasource, connector/action kept for the connector arm); cost: a published TS interface change and its consumers, about half a day on the spec lane", "B — fill connector with the datasource name and action with a fixed word for a datasource pull; cost: zero, but every reader of result.connector is told something false", "C — a second contract method for a datasource pull; cost: a new optional method, a second binder branch, more surface than the ADR names" ], "recommendation": "A, because the result is read by the binder and by any operator surface as 'where the rows came from'; B is a lie on a machine-readable surface and C doubles the door for one shape difference" }, { "question": "Lane conflict, blocks slices 3 and 6: ADR-0153's P2 engine row puts the _sync_ plan rows in packages/spec/src/data/injected-system-columns.ts (resolveInjectedSystemColumns) and rewrites JOB_ORGANIZATION_DESCRIPTION in packages/spec/src/system/job.zod.ts; the dispatch forbids packages/spec/src changes in this lane beyond the liveness rows. Not chosen silently.", "options": [ "A — the seat authorizes those two spec lines for this card (both are named by the accepted ADR row; the describe rewrite only restates D8's ruling)", "B — the spec lane carries them in a sibling PR that this card's slices then read" ], "recommendation": "A, because the ADR row assigns both to the engine child and a sibling PR adds a cross-lane ordering dependency for two small edits" }, { "question": "Where sys_sync_run is registered (D7, slice 2): beside SysAutomationRun in service-automation's plugin.ts registerRunObject (held by #32073), beside SysJob / SysJobRun in service-job's job-service-plugin.ts (the claim allows service-job only for D8), or as a platform-objects object registered by the runtime.", "options": [ "A — service-automation plugin.ts, after #32073 lands (the writer's own package; waits)", "B — service-job's plugin beside sys_job_run, the record it links (no wait; widens the claim's service-job use)" ], "recommendation": "A, because the pull executor is the one writer and the manual trigger needs no job service; the wait is one card" }, { "question": "Every over-cap read during halving is logged by the engine's find middleware as one WARN 'Find operation failed' with the refusal's message and stack (measured in this suite's output). A large source halved many times writes many such lines per run.", "options": [ "A — leave it; one summary info line per pull lands with D7", "B — the windowed read passes a read option that the engine's find honours by not warning on EXTERNAL_ROW_CAP_EXCEEDED (an engine option, objectql)" ], "recommendation": "A for now, because the lines carry no row data and an engine option is a wider change than the pull needs; revisit if an operator reports the noise" } ], "security_findings": "none", "out_of_scope_findings": [ "class: none (stale comment) · packages/spec/src/api/error-code-ledger.zod.ts still describes the EXTERNAL_ROW_CAP_EXCEEDED remedy as 'mirror the data with DataSync' in a code comment; not edited (no spec src change in this lane) · carrier: the slice or lane that next touches that file · noted, not filed" ] }11 remaining items
objectstack-fleet commentedon Nov 11, 2026
domain:server#7 · os-justin · session_01QSeKYmycB2eTbWXTULu4zX
os-dev-report addendum
PR #32385, patch round on the contract review 6786503520: S1, N2 and N3 are fixed, and N1, N4, N5 and N6 are left as directed. The head moved from
a87604ac(the seat's merge of main) tode2dd9b0. No merge of main was made in this round. The round is 10 files, +311 / −56.Changes
sys_sync_rungainsmapping_packageandtarget_object. The package is the resolved holder, never the caller:packageId, unless it is thesys_metadatastored-row marker or__runtime__. Core's resolution reads neither as a package.lastCommittedand the holder hand-over are scoped by holder package, mapping name and organization. The resume guard now also requires the same target object. The run is identified oncepullOnceresolves the mapping (SyncRunHandle.identify), anddescribestores the identity.jobs.mdx, the liveness container andwatermark.startnotes, and the object's doc and field descriptions.run_record_unavailablenames its four places inpull-refusal.tsand in the changeset: open, describe, the last-position read, and the commit. Each refusal sentence now says what was read. Thedescribeand last-position cases say "Nothing was read."; only the commit case says that the windows finished before stay written.organization_idcomment now says that undersinglethe engine stamps the Default Organization, which is why the ledger reads it back.check:livenessall pass unchanged.Pins. Five new real-engine cases in
datasource-pull.test.ts. The package cases resolve through the realmappingNameResolverover the engine's own registry.targetObjectwas edited starts its new target fromwatermark.start.Ablations on
de2dd9b0, each throughscripts/ablation-replace.mjs(anchor 1 → 0, restored blob == HEAD,git diff HEADempty), against the 34 tests ofdatasource-pull.test.ts:--holdand then the second mutation): 3 failed (cases 1, 3 and 4). Case 3 is held by either key part alone, so it reddens only with both off.Readings on
de2dd9b0check:test-typecheckOK.pnpm test: Test Files 275 passed (275), Tests 3733 passed (3733).datasource-pull,plugin-suspended-run-wiringandconnector-pull: 92 / 92.dispatch-gates.mjs --commandson this round's 10 paths derived 101 families, and all 101 ran: 100 exit 0, pluscheck-empty-changesetexit 1 (the declared DELIBERATE CORRECTION).check:dual-build-cjs-loadsfirst exited 3, with PREREQUISITE NOT MET, and exited 0 after the builds it reads.--ran: "101 derived famil(ies) accounted for — 101 run, 0 NOT-MEASURED (a DERIVED zero — all 101 recorded an exit code and none of them is 3)".The PR body is now wrong in these places (the dev does not edit it):
watermark.field": it is now per holder package and name, and the target must match too;run_record_unavailablebullet lists three places, not four;.changeset/32062-sync-in-datasource-pull.mdchanged again. The note now reads: "Where a run starts. Each run reads on from the read position the last run of the same mapping (its package and name) committed in the same organization, stored onsys_sync_run(ADR-0153 D7, in the same release: see its own entry), so it reads only the rows written since, plus the rows sharing that position, which the keyed upsert writes again harmlessly. The first run, every run of a mapping that declaresfull: true, and a run whose mapping now reads another source object or field or writes another target, read fromwatermark.start.deletePolicyanddeletedFieldstill warn their author that they change nothing yet."Counts: this comment is the round's one REST write, through
post-stamped.mjs.git pushis not a REST write, and the head-sha reply goes by session message. MCP calls: 0.objectstack-fleet commentedon Nov 11, 2026
domain:server#7 · os-justin · session_01QSeKYm8sWjp2Pk76warN7w
Slice 3 ACCEPT → PR #32385 comment 6787577456 · 2026-11-11T05:06Z.
objectstack-fleet commentedon Nov 11, 2026
domain:server#7 · os-justin · session_01QSeKYm8sWjp2Pk76warN7w
Slice 3 landed · next: slice 4 = the run's job id and trigger · 2026-11-11T05:34Z
PR #32385 merged as
ed95466e83(Part of; this card stays open and claimed). Every pull leaves asys_sync_runrow. A datasource pull reads on from the position the last run of the same mapping committed, keyed by holder package, mapping name and organization, and guarded by source, field and target. Contract review: PASS 6786503520, then delta PASS 6787442175. The DELIBERATE CORRECTION of the slice-2 changeset was confirmed in ACCEPT 6787577456.Claim widening for slice 4 (option A of the slice-3 open question). D7 lists the
sys_job_runid that drove a pull and its trigger. Scope:packages/spec/src/contracts/job-service.ts:JobHandler's context gains optionalrunIdandtrigger(schedule/manual). This is acontracts/hunk, a piggyback item per the spec lane,Clause-②: yes (widening).packages/services/service-job:DbJobAdapterpasses thesys_job_runid it already holds, and the cron and interval adapters pass the trigger. These are those adapters' handler-invocation lines only.packages/runtime/src/app-artifact-handlers.ts): forwarding lines only. runtime: a top-level job that shares its name with one package body's job is pulled as that body's package #32355 (this seat, in flight) editsjobDeclaringPackagesin the same file, so the slice-4 dev reads that branch before editing it, and mergesmainafter runtime: a top-level job that shares its name with one package body's job is pulled as that body's package #32355 lands if it lands first.ConnectorSourcePullRequestgains optionaljobRunIdandtrigger;sys_sync_rungainsjob_run_id(text, not a lookup, so job-ledger retention never cascades) andtrigger. Plus their tests,jobs.mdxand a changeset.Still after it, in the dev's order: D3 deletes, D4 provenance, D6 drift, the D8 refusals (
job_organization_missing), slice-1 N2, carrier note 6762989212 item 6, N7, and the per-dialect cases.objectstack-fleet commentedon Nov 11, 2026
domain:server#7 · os-justin · session_01QSeKYmycB2eTbWXTULu4zX
os-dev-report
{ "issue": 32062, "status": "done", "slice": "4 = ADR-0153 D7's job run id and trigger, scoped and claim-widened by the seat in 6788004425 (option A of the slice-3 open question); Part of #32062, the card stays open", "branch": "claude/issue-32062-sync-in-p2-engine", "pr": "https://github.com/objectstack-ai/objectstack/pull/32405", "session": "session_01QSeKYmycB2eTbWXTULu4zX", "premise_still_valid": true, "summary": "Contract: JobHandler's context gains optional runId and trigger (schedule / manual), with a new JobRunTrigger type; ConnectorSourcePullRequest gains optional jobRunId and trigger. Adapters: CronJobAdapter and IntervalJobAdapter tell the handler the trigger on each path (scheduled fire, with and without the cluster lock, and trigger()); DbJobAdapter adds the sys_job_run id of the attempt it is recording, one per attempt, and none under recordRuns: false. The runtime binder's pull form forwards both, each only as the contract spells it (two lines plus the run signature). sys_sync_run gains job_run_id (text, not a lookup) and trigger, written at open; a direct pull leaves both NULL. Deviations from the order: (1) the branch is this session's platform-assigned claude/issue-32062-sync-in-p2-engine, recreated from origin/main ed95466e (#32385 merged, its branch deleted on the remote), not claude/issue-32062-sync-in-p2-slice4, which this session may not push to; (2) the changeset level is patch, not minor: the level gate refuses minor beside Metadata-protocol: none, and no authorable key moves. packages/spec/src hunks beyond job-service.ts: contracts/automation-service.ts (ConnectorSourcePullRequest, the symbol the order names), and the two contract test files whose pins the widening moves (the four-keys pin becomes six; a new type-level block for the context); api-surface/ and export-origins/ record JobRunTrigger. One runtime test pin moved: app-plugin.job-data-reach.test.ts names the contract's trigger in the handler form's exact key set. DELIBERATE CORRECTION of the unreleased .changeset/32062-sync-in-run-record.md bullet that listed the job run and trigger as not recorded (quoted in the PR body). #32355 has not landed; the binder hunk sits apart from its hunks and a trial merge with its head ae41bb74 is clean, as is one with origin/main 767f1106.", "tests": "service-job: pnpm test, 12/12 files, 145/145 tests (94ed0284). New job-run-context.test.ts 9/9. runtime: the nine job and binder files, 127/127 (94ed0284). New app-artifact-handlers.job-pull-run.test.ts 6/6, including the real DbJobAdapter driving the real binder: a manual run and a scheduled fire each hand the pull the sys_job_run row the adapter wrote, with their trigger. service-automation: pnpm test, 275/275 files, 3737/3737 tests (da8c3e05; the package is unchanged after it); 4 new real-engine cases in datasource-pull.test.ts (job-driven, refused-at-open, trigger alone, direct pull NULL). spec: contract tests 37/37; check:generated, all 14 current with check:test-typecheck (e7e9bf94, 8192 heap). Typecheck exit 0 for service-job, service-automation and runtime, with check:test-typecheck OK. Runtime's first run refused one file for an unbuilt @objectstack/driver-turso (TS2307) and passed after it was built. Ablations: 17 on 0ba54a17, one per new forwarding line plus the two contract-spelling guards, each red, each restored byte-exact (blob == HEAD, git diff HEAD empty): cron 3/1/1/1, interval 5/3/2, DbJobAdapter 3, binder 3/4/1/1, ledger 2/3, pull 3/2/3 failed. The first run on da8c3e05 gave the same counts.", "gates": "dispatch-gates --commands derived 124 families from 22 paths at 0ba54a17; all 124 ran there: 123 exit 0, check-empty-changeset exit 1 (the declared DELIBERATE CORRECTION). --ran: '124 derived famil(ies) accounted for — 124 run, 0 NOT-MEASURED (a DERIVED zero — all 124 recorded an exit code and none of them is 3)'. Found on earlier heads and fixed: api-surface and export-origins (JobRunTrigger); objectql-double-limit and where-matcher (the new job-store doubles); engine-double-contract (update via assertEngineUpdateDispatch, then --write); test-source-alias (collection-time import in the edited suite); check:generated's test-typecheck leg (the four-keys pin). Prerequisite exits (3) for check:skill-examples and check:dual-build-cjs-loads cleared after building the fresh worktree. Level axis offline with the PR body as the event: exit 0, patch. A container restart stopped one run part-way; the final head's run is complete. CI check-runs: not yet read; the seat moves the head for CI if the push started none. Not run locally: pnpm lint and the CI farm.", "dogfood_battery": "n/a — no `!` in the PR title and the Clause-② line carries the widening arm", "mcp_calls": "3 — mcp__github__issue_read get #32062, mcp__github__pull_request_read get #32385, mcp__github__issue_read get_comments #32062 (the seat's 6788004425); no write tool", "api_writes": "2 — the relay pr_create (POST /repos/objectstack-ai/objectstack/dispatches, run 42986524201, read back identical) and this comment through post-stamped.mjs; git push is not a REST write", "open_questions": [], "security_findings": "none", "out_of_scope_findings": [ "class: b · reach: named producer DbJobAdapter (the default durable adapter): schedule() wraps the handler with a fixed defaultTrigger 'schedule' (wrap(name, handler, 'schedule', options)), and trigger() runs that same wrapper through the inner adapter, so a manual run's sys_job_run row reads trigger: 'schedule'; replay() adds its synthetic 'replay' row beside the wrapper's own 'schedule' row · evidence: db-job-adapter.test.ts:302 expects trigger: 'schedule' after adapter.trigger('on'); contradicts content/docs/automation/jobs.mdx ('trigger (schedule | manual | replay)') and the SysJobRun docblock ('whether the run was a manual trigger or a scheduled tick'); since this slice, a manual pull's sys_sync_run.trigger reads manual while the sys_job_run row it names reads schedule · dedupe words: sys_job_run trigger manual, DbJobAdapter defaultTrigger, manual trigger recorded schedule, replay row trigger · carrier: none — for the seat to file", "class: none (observation) · @objectstack/runtime's JobHandlerContext type does not declare runId or trigger, though the handler form passes the job service's context through at run time (pinned in app-plugin.job-data-reach.test.ts) · carrier: none · noted, not filed" ] }objectstack-fleet commentedon Nov 11, 2026
domain:server#7 · os-justin · session_01QSeKYm8sWjp2Pk76warN7w
Slice 4 ACCEPT → PR #32405 comment 6791958834 · 2026-11-11T09:17Z. Nits N1–N5 of record 6791935060 ride slice 5 (D3).
objectstack-fleet commentedon Nov 11, 2026
domain:server#7 · os-justin · session_01QSeKYm8sWjp2Pk76warN7w
Slice 4 landed · 2026-11-11T09:51Z
PR #32405 merged as
b96d492db8(Part of). This card stays open and claimed.Verified on
origin/main:JobHandler's context carries optionalrunIdandtrigger(JobRunTrigger).sys_sync_run(job_run_id,trigger).Next: slice 5 (D3, deletes). It carries nits N1–N5 of record 6791935060. It is not dispatched yet: it waits on the maintainer's decision about this card's seat, now that this seat is moving to the ADR-0150 epic seat.
Still after it, in the dev's order: D4 provenance, D6 drift, the D8 refusals, slice-1 N2, carrier note 6762989212 item 6, N7, and the per-dialect cases.
Related card: #32406 (
sys_job_run.triggerfor manual and replay runs).objectstack-fleet commentedon Nov 11, 2026
domain:server#7 · os-justin · session_01QSeKYm8sWjp2Pk76warN7w
Release: session_01QSeKYm8sWjp2Pk76warN7w (domain:server#7) · reason: the seat moves to the ADR-0150 epic seat on the maintainer's word, and he chose to send this card back to the server lane (「A」) rather than keep it as an exception · next:
pm:queuefor the nextdomain:serverseatResume-from: none
Written 2026-11-11T12:18Z. The slice branch
claude/issue-32062-sync-in-p2-enginewas deleted when its last PR merged. The next claimant starts a fresh branch fromorigin/main.State at release (
origin/main7d82b2dab8):datasourceSourcesys_sync_run, the committed read position (keyed by holder package, mapping name and organization; guarded by source, field and target), retention, boot closerunId/triggerreach the pull andsys_sync_run(job_run_id,trigger)Next: slice 5 = D3 (deletes). It carries nits N1–N5 of contract record 6791935060:
JobHandlerContext, the changeset and thejobs.mdxmember table should name thetrigger/runIdthe handler form now receives.originOfshould bound its inputs.jobs.mdxsentence about the two columns needs rewording. The replacement wording is in the record.sys_sync_runfield descriptions should say whenjob_run_idis empty.After it, in the dev's order:
injected-system-columns.tsplan rows;job_organization_missing;Open cards around it:
DbJobAdapterwrites a manualtrigger()and areplay()as'schedule'onsys_job_run. If its fix addsreplaytoJobRunTrigger, thesys_sync_run.triggerselect list moves with it.Release order: slices 1–4 are on
mainand unreleased. No release constraint is open between them; slice 3's S1 was fixed before landing.Claim surface for the next claimant: read the original claim 6769041777 and the slice-4 widening 6788004425. The
contracts/piggyback andservice-jobhandler lines were slice 4's and are done. D8 reopensservice-jobfor the organization predicate.Dev context: the cloud dev session that built slices 1–4 is told to stand down. Its knowledge is in the four PR bodies and the reports on this card.
objectstack-fleet commentedon Nov 13, 2026
triage#1 · hotlong · session_01Q4EjjMnyLqJdzuktEmRPXH
Path: none
Triage: regrade p1 → p2 under the adopted grading rules (#32997) · 2026-11-13T03:10Z
⛔ Not a claim; the card's claim and state are unchanged.
Generated by Claude Code