ci(temporal-conformance): give the non-SQL step the job's live PostgreSQL URL

[objectstack-fleet]

Fixes #21564
Clause-②: no

What this changes

The Temporal Conformance (live PG + MySQL) job's step "Run the non-SQL temporal backends under the skewed process zone" now sets OS_TEST_POSTGRES_URL in its env:, with the same literal the job's three SQL steps use (postgres://postgres:postgres@127.0.0.1:5432/postgres), for the postgres service the job already provisions. Until now that step set only TZ, so every service-analytics live-PostgreSQL cell (describe.skipIf(!config)) was a named skip there while the job read green.

The header clause "no CI step provisions that variable for this package" in 14 service-analytics test files becomes false with this wiring. It is corrected, comment-only, to name the step that now sets it. No other line of those files changes (every changed line in them is a * comment line).

No other job, no other step, and no timeout changes. No cell is re-skipped.

What the step needed, measured (A2)

Only the one env: line:

• Build closure. turbo run build --filter=@objectstack/service-analytics... --dry=json selects 16 packages, including @objectstack/driver-sql, @objectstack/objectql and @objectstack/driver-sqlite-wasm (dev dependencies of service-analytics). The step's existing build already builds what the cells import.
• PostgreSQL client. pg is an optional peer of driver-sql, resolved by pnpm into driver-sql's knex instance (knex@3.3.0_..._pg@8.22.0); require.resolve('pg') from that knex answers pg@8.22.0. The cells reach pg through SqlDriver, so nothing is added to service-analytics.
• Database and tables. Every cell connects to the default postgres database and creates and drops its own tables. The table names are distinct across the 16 files (os21316_deal, os21505_coercion, ...), so parallel vitest workers do not collide.
• Server zone. The job's existing step "Point both servers at a non-UTC timezone" sets the server to Asia/Shanghai before any test step.

Readings

• A3. No service-analytics file reads OS_EXPECT_LIVE_DIALECT_MATRIX (git grep exit 1), and no code in the step's other four packages reads it (the only hit, in driver-mongodb/src/test-mongod.ts, is a comment). So the flag is not added to the step, and no code reading it is added.
• A4. No service-analytics file reads OS_TEST_MYSQL_URL or any other server variable the job provides. Its only process.env reads are OS_TEST_POSTGRES_URL (32), NODE_ENV and TZ. Nothing is listed for MySQL.
• A5. core, formula, driver-memory and driver-mongodb have zero OS_TEST_POSTGRES_URL hits (git grep exit 1 for each). Adding the variable to the step starts no test in them.

Local run against a non-UTC PostgreSQL 16

PostgreSQL 16.14 started privately on a random port with timezone=Asia/Shanghai, process TZ=America/New_York (the job's two zones), service-analytics full suite, tree 3f9dcd91c4. Counts are passed / failed / skipped per file, from vitest's JSON reporter.

file	without the URL	with the URL
analytics-faces-one-lowering	69 / 0 / 20	88 / 0 / 1
cube-measure-field-type-door	12 / 0 / 11	23 / 0 / 0
cube-measure-relationship-path-type	7 / 0 / 6	13 / 0 / 0
json-stored-door-live-drivers	5 / 0 / 5	10 / 0 / 0
json-stored-door-undeclared-join	5 / 0 / 5	10 / 0 / 0
list-at-scalar-operator-both-faces	43 / 0 / 43	86 / 0 / 0
native-sql-aggregate-policies	25 / 0 / 24	49 / 0 / 0
native-sql-base-column-qualify	6 / 0 / 6	12 / 0 / 0
native-sql-boolean-comparand-door	34 / 0 / 34	68 / 0 / 0
native-sql-measure-number-presentation	5 / 0 / 5	10 / 0 / 0
native-sql-number-comparand-door	58 / 0 / 58	116 / 0 / 0
native-sql-offset-only-window	8 / 0 / 6	14 / 0 / 0
objectql-echo-offset-only-window	3 / 0 / 3	6 / 0 / 0
objectql-face-order-limit	13 / 0 / 13	22 / 4 / 0
order-key-selected	8 / 0 / 8	16 / 0 / 0
read-scope-temporal-coercion	10 / 0 / 6	15 / 0 / 1
16 files	311 / 0 / 253	558 / 4 / 2

Whole package: 4152 passed, 253 skipped without the URL; 4399 passed, 4 failed, 2 skipped with it. The two skips left are cell-scoped by design: the SQLite cell's "the server is not on UTC" check in read-scope-temporal-coercion, and the PostgreSQL cell's SQLite-only "a host with no typed reader" case in analytics-faces-one-lowering.

One live cell is red once wired

objectql-face-order-limit.test.ts, the live postgres cell, 4 tests:

test	assertion	expected	received
the card's row: ordered desc and limited to 1	:244 (native face)	[['2026-06', 1]]	[['2026-05', 2]]
ordered desc with offset 1 and limit 2	:257	[['2026-05', 20], ['2026-04', 1]]	[['2026-03', 6], ['2026-02', 10]]
CONTROL with no order, offset or limit, bucketed time dimension	:334	4 buckets	3
the dataset door, month-bucketed, desc, offset 1, limit 2	:363 (native face)	[['2026-05', 20], ['2026-04', 1]]	[['2026-03', 6], ['2026-02', 10]]

Every closed_on (a date field) lands one day early: 03-01 becomes Feb, 06-01 joins May.

• The axis is the server zone, not the process zone. The same file with the session pinned to UTC (options=-c TimeZone=UTC on the URL) and TZ=America/New_York: 26 passed. With the server on Asia/Shanghai and TZ=UTC: the same 4 failed.
• Where it comes from (a reading, not a fix). driver-sql's buildDateBucketExpr PostgreSQL branch (sql-driver.ts:6134-6138) buckets with to_char((col)::timestamptz AT TIME ZONE 'UTC', 'YYYY-MM'). On a date column, ::timestamptz resolves midnight in the session zone. In psql on the same server, to_char(('2026-06-01'::date)::timestamptz AT TIME ZONE 'UTC', 'YYYY-MM-DD') answers 2026-05-31 under Asia/Shanghai and 2026-06-01 under UTC.

Per the triage ruling, this is a real red. It is ⛔ not fixed here and ⛔ not re-skipped. The PM cards it to the cell's lane. Until then this PR's Temporal Conformance check is expected to be red, and the PR stays a draft.

Acceptance is read from this PR's own Temporal Conformance (live PG + MySQL) job log (cells run, not skipped), not from green. That reading, against a recent main run of the same step, is recorded in the os-dev report on the card.

Acceptance notes

• The ruling says each of these cells "asserts a non-UTC server". At this tree, 1 of the 16 files does (read-scope-temporal-coercion.test.ts:233-237, show timezone). The other 15 rely on the job's provisioning step for the server's zone. The red cell above is zone-sensitive and asserts no server zone. Noted, not changed.
• Stall guard. check:stall-guard-budget reads this step's static budget as window 10m, cap 20m, job budget 30m, slack 10m (2 of 2 guarded steps share the clock). The step's measured time on this PR is in the card report.
• Release. Nothing publishes: .github/workflows/ci.yml, and comments in test files outside service-analytics's files[] (dist, README.md, CHANGELOG.md). Measured after a build: os21316_deal, no CI step and the new step name have 0 hits in those files; positive control AnalyticsServicePlugin has 4. No changeset; skip-changeset.
• Governed. check-governed-merges.mjs --branch: 0 of 15 paths hit the register, NOT governed; 110 changed lines.

Gates

Run on head a3fb6c0334 (the merge of origin/main a3873542ca). Each exit code was recorded before any pipe.

• 85 of the 87 commands dispatch-gates.mjs --commands derives exited 0. This includes check:required-contexts, check-ci-filter-parity and its --self-test, check:workflow-step-name-quoting, check:workflow-status-functions, check:stall-guard-budget, check:stall-guard-headroom, check:pnpm-filter-targets, check:nul-bytes, check:changeset-gate-self-tests and check-changeset-no-major.
• check:dual-build-cjs-loads: exit 3, PREREQUISITE NOT MET. It needs a full pnpm build, and this diff changes no package's built output. NOT MEASURED; CI runs it.
• check:pm-dispatch-gates and check:type-check-debt: each ran past the local 420 s bound (exit 124) on the shared box, then were re-run without that bound. That result is in the card report.
• service-analytics: typecheck passed; its full test suite is in the table above.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

Nothing in this diff resolved to a documentable surface (no symbol, route or SDK anchor derived from 0 changed package(s)), so this run has no opinion about the docs.

What this run could not see

• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 0 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 6dd99b82c38cd68b51c86241d53c5b7dda670a08 → packageMentionDocs.