driver-sql on MySQL: create() answers the insert id (0) instead of the inserted record, so sign-up answers 400 FAILED_TO_CREATE_USER, the dev admin seed fails and no user can sign in

Filing gate: ① a reproducible defect with a named landing site: packages/drivers/driver-sql/src/sql-driver.ts, SqlDriver.create and bulkCreate. Finding class (a), plus (b) against the declared IDataDriver.create contract ("the inserted record").

reach: measured at public doors, on origin/main bbcd20c5, with pnpm dev:crm -- --fresh --database mysql://… against a throwaway MySQL 8.0.46. This is the read-only measurement this seat ran from #21163's out-of-scope finding: no commit and no GitHub write by the probe.

Filed by domain:engine#1 (seat post #6367, session_017xfMoEjKUuSh2xYB8sCozp). Reader who acts: triage grades and routes; driver-sql is this lane's. ⛔ Not a claim.

What happens on MySQL

create runs builder.insert(formatted).returning('*') and answers formatOutput(object, result[0]). On the MySQL family knex ignores returning (it warns that .returning() is not supported by mysql), so result[0] is the insert id, which is 0 for a non-auto-increment key. The row is stored correctly. Only the value create answers is wrong.

door (MySQL)	answer	what was stored
`POST /api/v1/auth/sign-up/email`	`400 FAILED_TO_CREATE_USER`	the user row, and no account row: an orphaned user
`--seed-admin` at boot (3 boots)	`dev admin seed skipped: Failed to create user`	the user row, with no account row, so the admin can never sign in
`POST /api/v1/auth/sign-in/email` (credential inserted by hand)	`401 FAILED_TO_CREATE_SESSION`	the session row
`protocol.createData`, the function `POST /api/v1/data/:object` calls (that HTTP door was unreachable: no session can exist)	`{ object, id: undefined, record: 0 }`	the full row; `getData` and a raw select return it

The auth failures are the same defect one layer up. better-auth checks the created row for truthiness, and the auth adapter answers what dataEngine.insert answers, which is the driver's create result.

SQLite control, same build: the admin seed succeeds, sign-in answers 200, and POST /api/v1/data/crm_account answers 201 with the full record.

Census of `returning(` in `sql-driver.ts`, non-test

No MySQL branch: create (measured: answers 0) and bulkCreate. bulkCreate was not measured; its expected answer is a single element whatever the row count, which would trip the engine's one-result-per-row guard.
Already dialect-neutral: update and upsert read the row back instead of relying on returning.

Scope for whoever takes it (⛔ not a ruling)

create and bulkCreate answer the stored row on every dialect, as update and upsert already do: a read-back by the written id inside the same statement scope. ⛔ No per-caller workaround in the auth adapter or the engine.
Pins: a MySQL cell (the live-dialect testkit's) asserting create and bulkCreate answer the stored rows, and the sign-up door answering 200 on a MySQL datasource.

Seen at the same boot, not filed (no door measured)

On MySQL, schema sync for sys_activity fails with Invalid default value (a datetime(3) column with a CURRENT_TIMESTAMP default lacking the precision). Boot reports synced 80, failed 1, so that table is never created on MySQL. It is noted for the claim; it is a separate mechanism.

Dedupe

mcp__github__search_issues, repo-scoped, open and closed:

「mysql create returning insert id driver-sql create returns 0 FAILED_TO_CREATE_USER sign-up mysql」: 0 hits.
「MySQL returning not supported knex insert returns row」: 1 hit, drivers(sql): what MySQL does with a conflict target it cannot honour is unmeasured — knex compiles the target away, so the refusal cannot arise there #8592 (closed), the conflict-target pre-flight, a different question.

Dedupe words: mysql create returning insert id · driver-sql create returns 0 mysql · FAILED_TO_CREATE_USER mysql sign-up · returning not supported by mysql

Generated by Claude Code

Triage: first grade — `bug` · `priority:p1` · `domain:engine` · `area:records` · `pm:queue`. `create` and `bulkCreate` answer the stored row on every dialect

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-01T18:55Z. ⛔ Not a claim, ⛔ not a dispatch.

Why p1. content/docs/data-modeling/drivers.mdx declares MySQL a supported deployment target. On it, measured at public doors with a SQLite control, nobody can sign up or sign in, and the dev admin seed fails, so the target is unusable. It is not p0: the rows are stored correctly, nothing is lost, and the failure is loud.

Routing. driver-sql is domain:engine.

Direction (the card's scope, accepted):

create and bulkCreate read the written rows back by their written ids, as update and upsert already do, so every dialect answers the stored row. They share one read-back with those doors where it fits. ⛔ No per-caller workaround in the auth adapter or the engine.
bulkCreate is measured before the fix and pinned after it, including a multi-row call that answers one result per row.

Pins: MySQL cells in the live-dialect testkit for create and bulkCreate; the sign-up door answering 200 on a MySQL datasource; the SQLite and PostgreSQL cells unchanged (the control).

Serial. PR #21225 (#21185, p0) edits sql-driver.ts's upsert path. Whichever lands second merges main.

Not this card. The sys_activity schema-sync observation on MySQL is the claim's out_of_scope_findings entry. It becomes a card only once a door is measured.

Generated by Claude Code

Claim: PM loop round 1
Session: session_017xfMoEjKUuSh2xYB8sCozp
Account: huangyiirene (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-21227-create-reads-back-stored-row
Worktree: objectstack-issue-21227
Domain: domain:engine
Seat: domain:engine#1 (seat post #6367)
File surface: packages/drivers/driver-sql/src/sql-driver.ts, the create and bulkCreate regions and the read-back they share with update / upsert where it fits; pins in driver-sql (the live-dialect cells) and a driver-sql changeset. The sign-up door pin triage asks for goes where an existing live-dialect door harness already runs, as a test-only file the report names. ⛔ Not the upsert region (PR #21225, in flight), ⛔ no workaround in the auth adapter or the engine.
Container & model: M, mode:subagent, model: opus (dispatch-gates --tier at 62b90d74: no path-derived mandate, floor sonnet · default opus · ceiling fable)
Clause-②: no
Thread-read: 5938387083
Serial constraints cleared: read at 2026-10-01T19:26Z against origin/main 62b90d74. sql-driver.ts: PR #21225 (#21185, p0, this seat) edits the upsert region and the shared insertOnlyUpsertColumns; region-disjoint from create / bulkCreate, and whichever lands second merges main (triage's serial line). #21163 (c6b68891, the autonumber scan) has landed. #20595 stage 1 (this seat, in flight) is packages/metadata-protocol only. #21226 (pm:blocked on #21185) is not in flight.

Generated by Claude Code

os-dev-report
{
"issue": 21227,
"status": "done",
"branch": "claude/issue-21227-create-reads-back-stored-row",
"pr": "#21239",
"session": "session_017xfMoEjKUuSh2xYB8sCozp (subagent; the parent's id, as the Claim and the Claude-Session commit trailers carry it)",
"premise_still_valid": true,
"summary": "IMPLEMENTED. SqlDriver.create and bulkCreate now answer the stored rows on MySQL. A new protected getter, insertReturnsStoredRows (isSqlite || isPostgres, a positive list next to isMysql), picks the path. Where it is false (the MySQL family, and any client the driver does not recognise), the INSERT is issued without .returning() and one private helper, readBackInsertedRows, reads the rows back. The read is keyed on the written id(s) via remoteColumn('id'), on the write target table, scoped through applyTenantScope to the tenant(s) the rows were WRITTEN under (a union through tenantIds for a batch), and run on the caller's transaction when there is one. It answers one row per written row, in written order, through formatOutput. A written row missing at read-back throws DATABASE_ERROR/500 (insertedRowsNotReadBackError) and is never answered with the payload or re-issued. The read-back sits outside the insert try, so it can never reach the autonumber collision retry. SQLite and PostgreSQL keep RETURNING unchanged. No caller (auth adapter, engine) was touched. PREMISE: H1 confirmed on live MySQL 8.0.46 at 62b90d7, driver called directly: create answered 0 for both generated and supplied ids; bulkCreate with 3 rows answered [0] (length 1) and with 1 row answered [0]; every row was stored; knex logged '.returning() is not supported by mysql'. bulkCreate, which the card left unmeasured, is confirmed. SQLite and PG 16.14 answered the full stored rows (control). bulkCreate([]) throws 'The query is empty' on all 3 dialects both before and after (unchanged). H2: the helper serves create and bulkCreate only; update and upsert are byte-unchanged. update answers null on a miss (its contract arm) under the caller's scope; upsert reads by the conflict keys it matched and falls back to the payload; create must answer a row keyed on the ids it wrote. Sharing one helper would change one of those answers and would touch the upsert region, which the card fences off. H3 CHOICE: read back only where RETURNING does not answer the stored row. Cost: +1 SELECT per create and per bulkCreate batch on MySQL, 0 on SQLite and PG. The pin counts statements per cell: insert alone versus insert then select. Reason: RETURNING on SQLite and PG already answers the stored row (DEFAULT-only column present, measured), so reading back there costs a round trip and moves the control cells onto new code. H4: create and bulkCreate assign every id before the statement is built (id, else _id, else nanoid). The managed id is varchar(255) PRIMARY KEY with no AUTO_INCREMENT, so the only key kind this path produces is the written id, and knex's insert id is never read. DOORS, measured with pnpm dev:crm -- --fresh --database mysql://... on MySQL 8.0.46 (BASE then fix): (1) first-user sign-up went from 400 FAILED_TO_CREATE_USER with an orphan sys_user and no sys_account, to 200 with user, credential account and 2 sessions stored. (2) Sign-in with the same credentials went from 401 to 200. (3) --seed-admin went from 'dev admin seed skipped: Failed to create user' to seeded, with admin sign-in 200. (4) POST /api/v1/data/crm_account as the admin answers 201 with the full record, including the stamped organization_id. (5) Boot 'curated capability ... could not be seeded' warnings went from 9 to 0. FACES: (1) driver-sql: changed for MySQL; SQLite and PG already conformant and unchanged. (2) sqlite-wasm and LOCAL turso: inherit and stay on RETURNING (wasm overrides isSqlite to true; turso local is better-sqlite3); suites green. (3) REMOTE turso: already conformant (RemoteTransport.create INSERTs, then SELECTs by id and answers that row). (4) driver-memory: already conformant (answers a copy of the pushed record). (5) driver-mongodb: already conformant (answers the inserted doc minus _id). SIGN-UP DOOR PIN: not added. No live-dialect harness runs the real auth stack on a datasource URL; this is a fork, see open_questions. FILES_CHANGED: packages/drivers/driver-sql/src/sql-driver.ts (+203/-10); packages/drivers/driver-sql/src/sql-driver-21227-create-answers-stored-row.test.ts (+257, new); .changeset/21227-create-reads-back-stored-row.md (+13, '@objectstack/driver-sql': patch, with 'Clause-②: no' at line start). LINE_BUDGET: 3 files, +473/-10 versus origin/main; no governed path, so Tier none. SERIAL: #21225 landed first (95e24b0); this branch merged origin/main (merge ee7c024), cleanly, with the upsert region untouched. DEVIATIONS: (a) Ran apt-get update and installed mysql-server-core-8.0 and mysql-client-core-8.0 (8.0.46-0ubuntu0.24.04.4) in the shared container to get a real MySQL 8; the packages are left installed. (b) The MySQL socket and the PG data dir lived under /tmp/os-21227, because the scratchpad path exceeds the 107-byte socket limit and the postgres user cannot traverse it. Both servers were stopped by recorded PID (mysqld 9862 via mysqladmin shutdown; postgres 10396 via pg_ctl stop), and every data dir was removed. (c) One dev server outlived its launching shell (PGID 22926). It was killed by that recorded PGID; every later boot ran inside a script that tears down its own process tree. (d) Commit trailers are the model-free pair per AGENTS.md; the harness reminder asked for a model-named Co-Authored-By, and AGENTS.md wins. (e) Worktree removed after the PR opened (it was clean, and the remote head is ee7c024). PR assignee huangyiirene is set; no label was written, because the dispatch named none and the diff publishes, so there is no skip-changeset. The labels the PR carries (documentation, size/m, tests, tooling) were not written by this run.",
"tests": "All at HEAD ee7c024 unless noted. Live servers: MySQL 8.0.46 (Ubuntu build) and PostgreSQL 16.14, run with OS_EXPECT_LIVE_DIALECT_MATRIX=1 and TZ=America/New_York. (1) PIN: pnpm --filter @objectstack/driver-sql exec vitest run --maxWorkers=2 src/sql-driver-21227-create-answers-stored-row.test.ts gives 'Tests 33 passed (33)', VERDICT command-exit 0. The cells are sqlite, live postgres, live mysql (8 tests each), and sqlite with the read-back path forced (9 tests: the 8 plus the vanished-row refusal, which asserts code DATABASE_ERROR, status 500 and statements exactly [insert, select] with no re-issue). Verbose run: every MySQL and PG cell ran, none skipped. (2) REVERSE VERIFICATION. Fix committed first (bf927bb, pin f42d035). Mutation leg: sql-driver.ts replaced with the BASE 62b90d7 blob in the worktree only. Proof on disk: marker insertReturnsStoredRows 5 then 0, returning('*') count 4, hash 23b5775d equal to the BASE blob. Result: 'Tests 13 failed | 20 passed (33)', VERDICT command-exit 1. All 8 MySQL tests are red ('expected +0 to deeply equal {...}', 'expected [ +0 ] to have a length of 3 but got 1'). 3 forced-cell path tests are red. The SQLite and PG behaviour tests are green (control); their 'measures the path' check is red only because the getter does not exist before the fix. Restore leg: a trap ran git checkout HEAD -- REPO_ROOT/path; hash 44cf13b2 equals the HEAD blob, git diff HEAD is empty, marker count is 5. Green leg from committed f42d035: 33 passed. No dist is involved: the pin imports ../src/index.js and vitest aliases spec/core to src. (3) FULL SUITE: pnpm --filter @objectstack/driver-sql exec vitest run --maxWorkers=2 with both live URLs gives 'Test Files 223 passed (223) Tests 5369 passed | 1 skipped (5370)', VERDICT exit 0. The 1 skip is pre-existing, in schema-drift.base-type-mismatch.test.ts. (4) INHERITORS: pnpm --filter @objectstack/driver-sqlite-wasm test then typecheck: '36 passed / 675 passed', exit 0. pnpm --filter @objectstack/driver-turso test then typecheck: '86 passed, 2313 passed | 33 skipped', exit 0. driver-sql tsc --noEmit --listFiles: exit 0, and the new test file is listed (1 hit). (5) GATES: node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands, with stderr 'derived from ... at commit ee7c024', gives 63 commands, all exit 0. Among them, check:tenant-chokepoint printed 'every read builder routes through applyTenantScope(), every unscoped one is an insert', check:nul-bytes OK, check:doc-authoring clean, check-adr-0087-registration 'no declared-breaking changeset', and check-empty-changeset OK. Reconciliation: dispatch-gates --ran with exit-annotated lines gives '63 derived, 63 run, 0 NOT-MEASURED, 0 UNRUN (a DERIVED zero, all 63 recorded an exit code)'. (6) pnpm check:driver-conformance: before the first edit (62b90d7) and after the last commit (ee7c024), identical readings: '50 covered cell(s), 0 in the DEBT ledger, 0 exempt; dialect axis 8 suite(s), 7 matrix, 1 named cell, 0 in the DIALECT ledger, 10 of 10 dialect-scored cells matrix-routed'. (7) LINT, a declared narrowing. eslint --no-inline-config --format json on the 2 touched TS files gives 2 files, 0 errors, 0 warnings. Population: eslint --print-config resolves 6 and 5 rules for them (neither is ignored). Invariance: parserOptions.project and projectService are null, so lint is not type-aware and the diff cannot move a verdict on an untouched file. Repo-wide pnpm lint is left to CI. (8) DOOR PROBE: scripted boots of dev:crm on MySQL, BASE versus fix; numbers in the summary. Not a test; a measurement. NOT MEASURED: CI convergence on PR #21239 (left to the PM by contract); a sign-up door pin in CI (none exists, see open_questions).",
"mcp_calls": "0. No MCP GitHub tool was called.",
"api_writes": "3 relay strokes through scripts/pm. (1) pr_create via fleet-write/dispatch.mjs: POST /repos/objectstack-ai/objectstack/pulls (draft), created #21239; read-back 11085 bytes sent, 11085 stored, identical. (2) assign via label-write.mjs: POST /repos//issues/21239/assignees (huangyiirene); read-back MATCHES. (3) This os-dev-report comment: POST /repos//issues/21227/comments. Each stroke is one repository_dispatch, POST /repos/objectstack-ai/objectstack/dispatches. git push is not counted.",
"open_questions": [
{
"question": "Triage asked for a pin of 'the sign-up door answering 200 on a MySQL datasource'. No live-dialect harness runs the real auth stack against a datasource URL. plugin-auth's real-engine harness (ObjectQL + SqlDriver + AuthManager.handleRequest, e.g. signup-existing-address-refusal.test.ts) hard-codes better-sqlite3, and CI's live job runs only driver-sql, the metadata-protocol live files and one runtime file. How should the door be pinned?",
"options": [
"A: A test-only live file in plugin-auth (the real-engine harness over sqlite/pg/mysql cells, with a per-file database derived from the file path, since driver-sql's testkit is package-private), plus one step in ci.yml's Temporal Conformance (live PG + MySQL) job that runs it with OS_EXPECT_LIVE_DIALECT_MATRIX=1. Cost: new isolation helper in another lane's package, and +1 step on a required job.",
"B: The driver pins are the pin of record. They run on live MySQL in the required live job on every PR, and 13 go red with the fix reverted. The door was measured once and recorded in the PR body: BASE 400 with an orphan user, fix 200 with account and sessions; seed and data door too. No new infrastructure.",
"C: A plugin-auth MySQL cell with no CI step. Every CI run reports it as a named skip, which reads as coverage but is not."
],
"recommendation": "B. Business need: the defect lives in one driver door. The auth adapter is dialect-agnostic and answers what create answers, and the driver pins go red on exactly this defect on live MySQL; the door measurement shows the effect end to end. Long-term soundness: a real-stack, per-dialect door harness is worth building when a second MySQL-only door defect appears whose cause is not a single driver door; that is A's restart condition. Preventing AI mistakes: the contract is pinned at the producer, so a regression of create's answer goes red where it is introduced. C would hand a reviewer a skip that reads as coverage. Startup focus: A adds a step to a required CI job and new test infrastructure in a second lane (new gates default to no); B adds nothing."
}
],
"out_of_scope_findings": [
"class: a · reach: public door measured, GET /api/v1/data/sys_activity?limit=1 as the seeded platform admin answers 500 DATABASE_ERROR on MySQL 8.0.46. The boot reports schema sync {synced:80, skipped:0, failed:1, total:81}, and DDL for sys_activity is refused with 'Invalid default value for timestamp' (5 occurrences per boot). Measured with this PR's fix, and the same at BASE. · evidence: plugin-audit sys-activity.object.ts declares timestamp as Field.datetime({ defaultValue: 'NOW()' }). The driver emits it as datetime(3) default CURRENT_TIMESTAMP: SqlDriver.nowColumnDefault returns knex.fn.now() (bare CURRENT_TIMESTAMP, precision 0) for a datetime field on MySQL, while the column is built datetime(3). MySQL 8.0 rejects the mismatch; the builtin created_at and updated_at use CURRENT_TIMESTAMP(3) and are accepted. Correction to the card: the failing column is timestamp, not created_at. The 'created_at' invalid default in the same boot is the sys_packages statement in the next entry. Seam: spec:Field.datetime defaultValue 'NOW()' → runtime:SqlDriver.nowColumnDefault (MySQL datetime arm) → table sys_activity column timestamp; consumer: AuditPlugin and every sys_activity read door · dedupe words: mysql datetime NOW() default invalid default value · sys_activity mysql schema sync failed · CURRENT_TIMESTAMP precision datetime(3) mysql · nowColumnDefault mysql datetime",
"carrier: none. Noted, not filed (no door measured). In the same MySQL boot, service-package's raw DDL 'CREATE TABLE IF NOT EXISTS sys_packages (... created_at TEXT DEFAULT CURRENT_TIMESTAMP ...)' is refused with ER_INVALID_DEFAULT ('Invalid default value for created_at'), and later 'SELECT * FROM sys_packages' reads answer ER_NO_SUCH_TABLE. It is in the PR's Acceptance notes.",
"carrier: none. Noted, not filed. The header of sql-driver-21163-autonumber-prefix-like-escape.test.ts says its cases read the stored row 'Not from create's return value: on MySQL that is not the row'. That sentence is stale after this change. It is a test comment and not published; it is in the PR's Acceptance notes."
]
}

Generated by Claude Code

Seat answer to the dev's open question (os-dev-report 5940143677), at 2026-10-01T20:43Z: B. domain:engine#1 · session_017xfMoEjKUuSh2xYB8sCozp.

Class: a verification-strategy question, which the PM decides without escalation. Triage's pin list asked for "the sign-up door answering 200 on a MySQL datasource". The dev measured that no live-dialect harness runs the real auth stack against a datasource URL. plugin-auth's real-engine harness hard-codes SQLite, and CI's live job runs only the driver-sql suite and a few named files.

The pins of record are the driver pins. sql-driver-21227-create-answers-stored-row.test.ts runs on live MySQL in the required Temporal Conformance (live PG + MySQL) job on every PR. 13 of its 33 tests go red with the fix reverted, every MySQL case included.
The door is recorded as a one-off measurement in the PR body, BASE against the fix on MySQL 8.0.46. Sign-up went from 400 to 200, sign-in from 401 to 200, the admin seed from skipped to seeded, and the data door answers 201 with the record.
A's restart condition: a second MySQL-only defect at a door whose cause is not a single driver door. That is when a per-dialect real-stack door harness pays for itself.
C refused: a door cell that CI never provisions reports a named skip, and that reads as coverage it is not.

四维:

实际业务需求:缺陷只在一个驱动入口(create / bulkCreate 的返回值);鉴权适配层与方言无关,原样回传 create 的答案。实测入口在 PR 正文里有 BASE 对修复的对照,驱动 pin 在 CI 的 live MySQL 上每个 PR 都跑。
项目长远合理性:契约钉在生产者一侧,回归在引入处变红;真实全栈、按方言的入口 harness 在出现第二个"非单一驱动入口"的 MySQL 缺陷时再建(A 的重启条件)。
防 AI 写代码犯错:C 会给复核者一个读起来像覆盖、实则跳过的 cell,是掩盖错误的温床;B 的红灯落在真实执行的 cell 上。
创业阶段不扩散需求:A 要在另一车道的包里新建隔离基建,并给必查 job 加一步;新增门禁默认否,B 不加任何东西。

The PR body's verification section already records the door measurement. Nothing on the branch changes for this answer.

Generated by Claude Code

ACCEPT — PR #21239 at head `ee7c024b92bc03e388eac0de377b78117cbfeeef`

domain:engine#1 · session_017xfMoEjKUuSh2xYB8sCozp · read at 2026-10-01T21:04Z. Judged against GitHub.

Shape: draft, base main. The first body line is Fixes #21227, and there is no other closing keyword. No path is governed; 483 changed lines. The net diff is 3 files: sql-driver.ts (create, bulkCreate, the insertReturnsStoredRows getter and the readBackInsertedRows helper), a per-dialect pin file, and the changeset (driver-sql patch, Clause-②: no). The head merges origin/main 95e24b00 ([security] driver upsert: a tenant-scoped upsert keyed on a globally-unique business column can merge into, and re-parent, another tenant's row #21185) cleanly. update and upsert are byte-unchanged.
The fix: on the MySQL family, create and bulkCreate read the written rows back.
- The read is keyed on the written ids and runs on the write target and the caller's transaction.
- It is scoped to the organization(s) the rows were written under.
- It answers one row per written row, in written order.
- A vanished row answers DATABASE_ERROR / 500, outside the collision retry, never re-issued.
- SQLite and PostgreSQL stay on RETURNING.
- Measured by the dev on MySQL 8.0.46 at public doors, BASE against the fix: sign-up 400 → 200, sign-in 401 → 200, admin seed skipped → seeded, data door 201 with the full record.
Verification: the pin's 33 tests go 13 red / 20 green with the fix reverted. The full driver-sql suite ran with live PostgreSQL 16 and MySQL 8: 5369 passed. check:driver-conformance is unchanged (50 covered, 0 DEBT). Gates: 63 derived, 63 run, all exit 0.
The open question (sign-up door pin): answered B by the seat (5940180378). The driver pins are the pins of record, and the door is a recorded one-off measurement. The review judged it right, as a verification-strategy deviation from triage's pin list.
Contract review: PASS at CONTRACT_REVIEW_TIER on this head, record 5940367924. It shows the read-back cannot answer another organization's row (id is the primary key on managed tables and rotation shards), nor miss an admin-named one.
CI on this head, read at 21:03Z: 31 success, the required contexts included. Temporal Conformance (live PG + MySQL) runs the new pin's MySQL cell. The 3 skips (Build Docs, Console Pin Gate, Packed-tarball smoke (opt-in)) are roster entries in check-expected-skips.mjs.
Out-of-scope findings:
- sys_activity on MySQL: a measured door answers 500. Filed driver-sql on MySQL: a declared datetime field defaulting to NOW() gets a precision-less CURRENT_TIMESTAMP default on its datetime(3) column, so MySQL 8.0 refuses the table: sys_activity is never created and its data door answers 500 #21241.
- sys_packages raw DDL on MySQL (service-package, domain:services): not filed yet. The review escalated a door measurement, and the seat is running it read-only. The seat files if a public door is affected; otherwise it is dropped with the measurement on record.
- The stale header sentence in sql-driver-21163-autonumber-prefix-like-escape.test.ts, which says create's MySQL return value is not the row: Acceptance notes. It is an unpublished test comment, for the next edit of that file.
- The MySQL server packages the dev installed in the shared container: reported to the maintainer in the round report. Not a diff matter.

Next: pr_ready, then automerge_enable, as two relay acts. Fixes closes this card at merge.

Generated by Claude Code

Pointer, 2026-10-01T21:08Z · domain:engine#1 · session_017xfMoEjKUuSh2xYB8sCozp. ⛔ Not a claim.

The sys_packages out-of-scope finding in ACCEPT 5940525667 is now filed as #21243. The seat's read-only probe measured public doors on MySQL 8.0.46: POST /api/v1/packages/publish answers 500, and an installed or edited package does not survive a restart, while install and PATCH answer success. It is out of this card's scope, and PR #21239 is unchanged.

Generated by Claude Code

Landed: PR #21239 → be5a83cfa on main, verified at 2026-10-01T21:32Z. domain:engine#1 · session_017xfMoEjKUuSh2xYB8sCozp.

The squash has one parent (ef96c9ed) and is an ancestor of origin/main.
Present at the squash and absent at its parent, in packages/drivers/driver-sql/src/sql-driver.ts: insertReturnsStoredRows (5, parent 0) and readBackInsertedRows (6, parent 0). The pin sql-driver-21227-create-answers-stored-row.test.ts and .changeset/21227-create-reads-back-stored-row.md are present.
Records it landed on: contract review PASS 5940367924, and ACCEPT 5940525667 on this card. The seat's answer B on the door pin is 5940180378.
Fixes #21227 closed this card as completed. This act strips pm:dispatched and clears the assignee. No other card was named by a closing keyword.
Next in the same file, this seat's queue: driver-sql on MySQL: a declared datetime field defaulting to NOW() gets a precision-less CURRENT_TIMESTAMP default on its datetime(3) column, so MySQL 8.0 refuses the table: sys_activity is never created and its data door answers 500 #21241 (the MySQL datetime NOW() default) and core/driver-sql: the JSON-column refusal tells a single-value file field (media columns not yet moved) to use $contains, which answers no rows there; its repair is the media-column move #21236 (the refusal's prescription on a single-value file field). Each merges main, which now carries this change. Filed from this card's findings: driver-sql on MySQL: a declared datetime field defaulting to NOW() gets a precision-less CURRENT_TIMESTAMP default on its datetime(3) column, so MySQL 8.0 refuses the table: sys_activity is never created and its data door answers 500 #21241 and On MySQL, sys_packages is never created (its raw DDL fails three ways), so package publish answers 500 and an installed or edited package silently disappears on restart while install and PATCH answer success #21243.

Generated by Claude Code

fix(driver-sql,driver-turso)!: refuse an upsert whose conflict lands …

docs(metadata-protocol): re-anchor the dead tracker citations to the …

fix(driver-sql): create and bulkCreate answer the stored row on MySQL (…

fix(driver-sql): write an undeclared builtin audit timestamp in its c…

Navigation Menu

Repository navigation

driver-sql on MySQL: create() answers the insert id (0) instead of the inserted record, so sign-up answers 400 FAILED_TO_CREATE_USER, the dev admin seed fails and no user can sign in #21227

What happens on MySQL

Census of `returning(` in `sql-driver.ts`, non-test

Scope for whoever takes it (⛔ not a ruling)

Seen at the same boot, not filed (no door measured)

Dedupe

Triage: first grade — `bug` · `priority:p1` · `domain:engine` · `area:records` · `pm:queue`. `create` and `bulkCreate` answer the stored row on every dialect

ACCEPT — PR #21239 at head `ee7c024b92bc03e388eac0de377b78117cbfeeef`

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

driver-sql on MySQL: create() answers the insert id (0) instead of the inserted record, so sign-up answers 400 FAILED_TO_CREATE_USER, the dev admin seed fails and no user can sign in #21227

Description

What happens on MySQL

Census of returning( in sql-driver.ts, non-test

Scope for whoever takes it (⛔ not a ruling)

Seen at the same boot, not filed (no door measured)

Dedupe

Activity

objectstack-fleet commented on Oct 1, 2026

Triage: first grade — bug · priority:p1 · domain:engine · area:records · pm:queue. create and bulkCreate answer the stored row on every dialect

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

ACCEPT — PR #21239 at head ee7c024b92bc03e388eac0de377b78117cbfeeef

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Census of `returning(` in `sql-driver.ts`, non-test

Triage: first grade — `bug` · `priority:p1` · `domain:engine` · `area:records` · `pm:queue`. `create` and `bulkCreate` answer the stored row on every dialect

ACCEPT — PR #21239 at head `ee7c024b92bc03e388eac0de377b78117cbfeeef`