fix(driver-sql,service-analytics): bucket the ISO week natively on SQLite, and the SQL echo refuses a bucket SQLite cannot run

[objectstack-fleet]

Fixes #21595
Clause-②: no

What changes

(a) week on SQLite, in driver-sql. The SQLite capability row now advertises week, and SqlDriver.buildDateBucketExpr's SQLite arm renders the ISO 8601 week as YYYY-Www, the label the PostgreSQL arm's IYYY"-W"IW and the MySQL arm's %x-W%v answer:

(strftime('%Y', ARG, '-3 days', 'weekday 4') || '-W' ||
 printf('%02d', (cast(strftime('%j', ARG, '-3 days', 'weekday 4') as integer) - 1) / 7 + 1))

It is the Thursday rule. '-3 days', 'weekday 4' moves any day to the Thursday of its Monday-start week, so that Thursday's %Y is the week-numbering year and its day of year gives the week number. It uses no %V or %G, and both date modifiers predate every SQLite the platform ships. ARG is the existing sqliteTemporalArg: a bare column for TEXT storage, and the legacy repair for an un-backfilled Field.datetime. No zone is read, so a Field.date stays its calendar day (H2 holds). Both callers follow without a second edit: aggregate()'s GROUP BY and the public dateBucketSql() the analytics echo reads.

(b) A non-UTC zone on SQLite, in service-analytics' echo. ObjectQLStrategy.generateSql no longer prints date_trunc(...) on a SQLite datasource, since SQLite has no such function. Where no driver expression answers and the executing driver's dialect (sqlDialectFor) is sqlite, the echo refuses with NOT_IMPLEMENTED / 501 and refusal: true. The shipped composition reaches that branch for a non-UTC timezone: the engine then buckets in memory on that zone's calendar, and SQLite has no time-zone database. The refusal reaches each door as follows:

• POST /api/v1/analytics/sql answers the refusal in the error envelope. refusal: true is the declared channel (ApiErrorSchema.refusal) for a 5xx whose prose is addressed to the caller, so the boundaries keep the message.
• POST /api/v1/analytics/query still answers the rows. execute() already calls generateSql inside a try, so the answer simply carries no sql. This is the reach the existing cross-field decline has.
• With timezone: 'UTC', or none, the echo still prints the driver's expression. On a dialect the host cannot name, it still prints the representative date_trunc.

Riders.

• R1: packages/objectql/src/engine.ts. The ADR-0053 D2 comment above tzRequiresInMemory now names driver-sql's to_char / date_format / strftime. In the same block, the comment's example "partial-support dialects (e.g. SQLite + week)" is false after this change; it now names the Turso remote transport, which advertises none. Comments only: 8 changed lines, 0 non-comment.
• R2: driver-turso's REMOTE_FACE_ANSWERS row dateBucketSql: 'inherited' is now pinned by running it. The remote face renders the expression with no connection, and the pin sends it through the remote face's own execute() to a real @libsql/client file: client, beside a better-sqlite3 local control. All five granularities answer the local labels, and week answers the written-out ISO weeks of the boundary days. The row's text needed no change (H4).

Comment corrections carried from #21485's ACCEPT 5971379160. objectql-strategy.ts, strategies/types.ts and the echo pin's docblock quoted PostgreSQL's to_char((col)::timestamptz AT TIME ZONE 'UTC', ...) as the bucket expression. Since 440cd32 that is the datetime expression only; each now names (col)::date::timestamp for a date.

Measured first: each SQLite client's bundled SQLite

client	package	sqlite_version()	strftime('%V', '2021-01-03')
better-sqlite3 13.0.3	driver-sql, driver-turso local face	3.53.4	53
@libsql/client 0.18.0	driver-turso remote face (file: client)	3.45.1	NULL
sql.js 1.14.2	driver-sqlite-wasm	3.49.1	53

The platform ships three SQLite clients, not two: SqliteWasmDriver is a SqlDriver subclass that overrides isSqlite, so it inherits this arm. libSQL bundles 3.45.1, so by triage's rule %V is out, and the expression is version-independent.

The expression, measured before it was written

• On each client: 12,454 inputs per client, with 0 mismatches against @objectstack/core's bucketDateKey on better-sqlite3, libSQL and sql.js alike. The inputs are every day from 1999-01-01 to 2032-12-31, datetimes at 00:00, 12:00 and 23:59:59.999Z on the ISO-year boundary days, the years 0001 / 0049 / 0050 / 0999 / 1000 / 9999, and unparseable and empty values, which answer NULL as the in-memory face does.
• Against PostgreSQL's own arm: PostgreSQL 16.14 to_char(d::date::timestamp, 'IYYY"-W"IW') over 1999-01-01..2032-12-31, 12,419 days, 0 mismatches. The boundary days: 2020-12-31 and 2021-01-03 are 2020-W53, 2021-01-04 is 2021-W01, 2024-12-30 is 2025-W01, 2026-12-28 and 2027-01-03 are 2026-W53.
• MySQL's %x-W%v: NOT MEASURED here, because no MySQL server is available in this container. It is pinned to the same bucketDateKey by CI's live Temporal Conformance.

(b): the declared response contract, read first

• /analytics/sql: AnalyticsSqlResponseSchema declares data: { sql: z.string(), params } over BaseResponseSchema, whose error is the refusal envelope. A null statement with a reason is not admitted, so it is not the shape, and the contract is not widened. A refusal is admitted, and this door already refuses a cross-field comparison (the [spec] service-analytics' read-scope / Cube filter compilers still refuse $field, so a CEL field-to-field RLS rule 400s on those faces #7598 ruling, "一致的响亮答案,不半渲染").
• /analytics/query: sql is z.string().optional(), "Generated SQL (if available)", so an answer without it is admitted.

So the shape is the refusal. The status is NOT_IMPLEMENTED / 501 rather than a 400: the query is spelled correctly and served, and the gap is the backend's. This is the reasoning driver-sql's own bucket refusal gives (#6212 / #5907), and the /meta/:type/:name/references 501 is the precedent for refusal: true.

Fixture triage: pins that held the removed branch

file	disposition
driver-sql/src/sql-driver-date-bucket.test.ts	caps.week is now true, and week joins the parity sweep. The #6212 refusal pin moves off SQLite week to the population that still reaches it: an unmodeled client (mssql, whose capability row is empty; the refusal is raised before any connection). Code, status and message as before, with the tail now Bucketed here: none (dialect 'mssql').
driver-sql/src/sql-driver-date-bucket-storage.test.ts	week joins both storage forms. One new case covers week over the legacy mixed-form column (epoch INTEGER, ISO TEXT, zone-naive TEXT, NULL), where the repair is expanded twice.
driver-sql/src/sql-driver-temporal-dialect.test.ts	week joins the dialect-gating granularities, so the legacy repair, the plain column form and the binding count are pinned for the new arm. New assertion: no SQLite arm reads a Field.date through 'localtime' or 'utc' (H2).
driver-sql/src/sql-driver-21485-date-bucket-calendar-day.test.ts	SQLite's BUCKETED_IN_SQL gains week, so aggregate() and the dateBucketSql echo run it on the date/datetime boundary rows.
driver-sql/src/sql-driver-13714-...test.ts	Comment only. The sweep is data-driven by the capability row and now serves week.
driver-sqlite-wasm/src/sqlite-wasm-driver-date-bucket.test.ts	caps.week is now true, and week joins the parity sweep on sql.js. Its refusal twin is removed: no granularity reaches that refusal on this driver any more.
driver-turso/src/remote-transport-groupby-node.test.ts	The #5240 parity case ("one condition, one wording") compares the remote transport with an unmodeled SqlDriver, the only SqlDriver face that still declines. The month asymmetry case is parameterized over month and week. The case that pinned the SQLite face's week refusal tail is removed.
qa/dogfood/test/date-bucket-parity-turso.test.ts	The tripwire made a local driver advertise week because SqlDriver could not run it. It now keeps the real row (which advertises week) and makes the week arm answer null, the incoherent pair a revert of the arm alone would ship. checkDateBucketParity still names it.
service-analytics/src/__tests__/objectql-echo-date-bucket.test.ts	The SQLite cell groups week. The in-memory-granularity FALLBACK case is removed, because its only input was SQLite week. The non-UTC case is per cell: on SQLite it pins the refusal (rows served with no sql; /analytics/sql refuses with NOT_IMPLEMENTED, 501 and refusal: true; declaredRefusalMessage(err) === err.message; a UTC control renders). On PostgreSQL it is unchanged (see the finding below). A unit case pins the refusal for a SQLite host that wires no hook.

Also swept and left alone: checkDateBucketParity's real-driver half in date-bucket-parity-conformance.test.ts now exercises week on better-sqlite3 and sql.js with no edit. Its week: false fake is self-contained. timedimension-granularity-driver-alias.test.ts is data-driven and green; see the Acceptance notes.

Reverse verification (commit first, then revert only the arm)

Taken at head 59a751f4ad, through scripts/ablation-replace.mjs (wrap mode: literal anchor, restore armed on exit, INT and TERM). The mutation replaced the whole SQLite week block with the original case 'week': return null; // see capabilities note. Anchor 1 to 0, blob 5f6b143dbe1a to 5e80a2795453. The capability row stayed week: true.

• Rebuilt @objectstack/driver-sql, then ablation-dist-preflight.mjs @objectstack/driver-sql "'-3 days', 'weekday 4'" --absent: exit 0, so the arm was out of the dist/ that driver-turso, driver-sqlite-wasm, service-analytics and dogfood consume.
• Red on every client (the predicted direction):
  • better-sqlite3 (driver-sql): 3 failed, the three week cases.
  • libSQL (driver-turso R2): 2 failed, week: libSQL answers the labels the local face answers and week: libSQL answers the ISO week of each boundary day.
  • sql.js (driver-sqlite-wasm): 1 failed, granularity=week.
  • The echo (service-analytics): 2 failed, closed_on, week and closed_at, week.
  • Dogfood conformance: 2 failed, both drivers.
• Restore: git checkout HEAD -- ABSOLUTE_PATH. The blob after restore is 5f6b143dbe1a, equal to HEAD, git diff HEAD is empty and git status --porcelain shows 0 entries. Then a rebuild and the presence preflight: the marker is in 2 built files and the tree reading is clean.

Tests

Package suites at head 9328c3c6cc (the merge of origin/main came after; its incoming commits touch none of these packages except driver-mongodb, where this diff is comments only):

• @objectstack/driver-sql, with the live PostgreSQL cell at TZ=America/New_York: 225 files passed, 3 skipped; 4571 tests passed, 104 skipped (the MySQL cells: no server here).
• @objectstack/service-analytics, with the live PostgreSQL cell: 176 files passed; 4421 tests passed, 2 skipped.
• @objectstack/driver-turso: 88 files passed; 2373 tests passed, 33 skipped.
• @objectstack/driver-sqlite-wasm: 36 files passed; 675 tests passed.
• typecheck: driver-sql, driver-sqlite-wasm, driver-turso, service-analytics, driver-mongodb (with check:test-typecheck) and objectql (all three tsc programs and check:test-typecheck) are all clean.

On the merged head 0e7e975a32, after a post-merge turbo build and spec check:generated (15 of 15 artifacts up to date), every changed pin passed again:

package	files	tests
driver-sql	5	99 passed, 2 skipped
driver-turso	2	109
driver-sqlite-wasm	1	8
service-analytics	2	31
dogfood	2	13
driver-mongodb mongodb-date-bucket-parity	1	39

Declared narrowings, so they read as such:

• objectql and driver-mongodb suites were not run in full. Their diffs are comment-only (engine.ts 8 changed lines, mongodb-aggregation.ts 14 and its test 4, all comment lines), and both packages typecheck.
• packages/qa/dogfood was not run in full; the two date-bucket files it changes were. The rest is CI's Dogfood Regression Gate.
• MySQL cells were not run.

The live PostgreSQL was a throwaway 16.14 instance this run started on a random port with server TimeZone Asia/Shanghai; it has been stopped and removed.

Gates

node scripts/pm/dispatch-gates.mjs --commands with no paths, at head 0e7e975a32: 101 derived (the 58 the dispatch named plus 43 the change set adds). All 101 ran on that head and each exited 0. --ran reconciliation: "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)". On the first run at 59a751f4ad, check:skill-examples and check:dual-build-cjs-loads answered exit 3 (PREREQUISITE NOT MET, unbuilt packages). After a full build both passed, and on the final head all 101 are exit 0.

Driver-card ledger, pnpm check:driver-conformance: before (at BASE 5c9138b4b6) and after (at 0e7e975a32), both 50 covered cells, 0 in the DEBT ledger, 0 exempt. No new DEBT.

Lint, a proven narrowing at 0e7e975a32:

• Population, read from eslint's own config: the 16 changed .ts files, run as pnpm lint runs (--no-inline-config). None was reported as ignored.
• --format json count: 16 files, 0 errors, 0 warnings.
• Invariance: eslint.config.mjs never enables type-aware linting (no parserOptions.project, no typed rules), so this diff cannot move the verdict on an untouched file.

File surface, and what this adds to the claim's

The claim named sql-driver.ts, the driver-turso pin, the service-analytics echo, its pin and two comment files, engine.ts R1, tests and the changeset. Each addition below is forced by this change, and none is a behaviour change outside the claim:

• Test files that pinned the removed branch (see the triage table): driver-sqlite-wasm, driver-turso's remote-transport-groupby-node.test.ts, and qa/dogfood's Turso parity tripwire.
• Statements this change makes false, each corrected as a sentence:
  • driver-mongodb's TSDoc on the exported MONGODB_DATE_GRANULARITIES ("driver-sql sets week: false", which ships in its .d.ts), and its test's comment.
  • content/docs/data-modeling/queries.mdx ("SQLite reports week: false, for instance").
  • The second engine.ts comment, in R1's own block.

Changeset .changeset/21595-sqlite-week-bucket.md: patch for @objectstack/driver-sql and @objectstack/service-analytics. driver-turso ships no code change, so it gets none, and the driver-mongodb and objectql changes are comments.

Acceptance notes

• Not filed, and left outside this diff:
  • service-analytics/src/__tests__/timedimension-granularity-driver-alias.test.ts's header still says SQLite declines week. The test is data-driven and green; the file is outside the declared cross-lane set.
  • The spec's queryDateGranularity TSDoc example ({ day: true, week: false } "for a SQLite build whose strftime lacks ISO week") is still a valid hypothetical, but no shipped SQLite driver declares it now.
  • objectql/src/engine-aggregate-timezone.test.ts's header names native bucketing date_trunc, R1's drift in a test file. None of these is made false by behaviour in this PR in a way any test reads.
• skills/objectstack-ui/rules/dashboards.md's week label row ("ISO date of the bucket (YYYY-MM-DD)"; the key is YYYY-Www) and skills/objectstack-query/rules/aggregation.md line 82 ("DATE_TRUNC etc.") were inaccurate before this change. They belong with skills(objectstack-ui): the dashboards rule says Postgres buckets with date_trunc; the SQL driver groups by to_char(... AT TIME ZONE UTC) on Postgres #21588, the open card for the same dashboards.md engine sentence, on a Tier H surface this PR does not touch.
• For the seat to file, (b)'s foreseen follow-up on PostgreSQL and MySQL. On PostgreSQL the echo for a non-UTC zone still prints date_trunc('month', closed_at); this PR leaves it, and the echo pin's PostgreSQL cell still asserts it. That statement was run on PostgreSQL 16.14 (server TimeZone Asia/Shanghai) over the pin's five rows:
  • It answers 2026-01-01 00:00:00+08-shaped keys on the session zone's calendar, where the face answers 2026-01 on the request zone's.
  • For timezone: 'America/New_York' the groupings differ too: the echo gives January 20 and February 8, where the face gives January 27 and February 1.
  • On MySQL date_trunc is not a function. That is NOT MEASURED here (no server).

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

This PR changes 4 package(s): @objectstack/driver-mongodb, @objectstack/driver-sql, @objectstack/objectql, @objectstack/service-analytics, touching 7 documentable anchor(s). ⚠️ 1 changed file(s) yielded no anchor (packages/drivers/driver-mongodb/src/mongodb-aggregation.ts), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files.

9 hand-written doc(s) NAME something this change touched and may need an implementation-accuracy re-verification:

• content/docs/data-modeling/drivers.mdx (via SqlDriver (symbol, a top-level class))
• content/docs/data-modeling/index.mdx (via SqlDriver (symbol, a top-level class))
• content/docs/data-modeling/queries.mdx (via SqlDriver (symbol, a top-level class))
• content/docs/permissions/tenant-audit-census.mdx (via SqlDriver (symbol, a top-level class))
• content/docs/plugins/packages.mdx (via SqlDriver (symbol, a top-level class))
• content/docs/protocol/kernel/index.mdx (via SqlDriver (symbol, a top-level class))
• content/docs/protocol/kernel/lifecycle.mdx (via SqlDriver (symbol, a top-level class))
• content/docs/protocol/objectql/query-syntax.mdx (via SqlDriver (symbol, a top-level class))
• content/docs/protocol/objectql/types.mdx (via SqlDriver (symbol, a top-level class))

⛔ 4 release-owned page(s) also name something this change touched. These are read-only:

• content/docs/releases/v14.mdx (via generateSql (symbol, a method of class ObjectQLStrategy))
• content/docs/releases/v17/17-0.mdx (via ObjectQLStrategy (symbol, a top-level class), SqlDriver (symbol, a top-level class))
• content/docs/releases/v17/17-5.mdx (via SqlDriver (symbol, a top-level class), generateSql (symbol, a method of class ObjectQLStrategy))
• content/docs/releases/v17/17-6.mdx (via SqlDriver (symbol, a top-level class))

content/docs/releases/ is RELEASE-OWNED (AGENTS.md "Documentation Guardrails"): release
notes are written centrally at release time, and a code PR that edits them is the exact PR
that guardrail exists to stop. They are still audited — read-only. If one of them is actually
wrong, file an issue or open a dedicated docs-only PR; do not edit it here.

What this run could not see

• 1 changed file(s) yielded no anchor (packages/drivers/driver-mongodb/src/mongodb-aggregation.ts) — pages documenting those are invisible to this run
• 1 name(s) were too generic to anchor anything (single lowercase words)
• the SDK route bridge reached 54 of 206 client-bound route-ledger rows — the other 152 have no registrar path: tail to select them, so pages documenting THEIR client methods cannot appear above, on this or any run. Of those 152: 0 are remediable by widening that discovery convention (an in-repo file declares the path; the convention did not scan it); 55 are structural — on a ledger where NOT ONE row is declared in-repo, so no discovery change reaches them at any price; 97 are undecided (no in-repo declaration, on a ledger that has other in-repo registrars — absence and an unreadable spelling are not distinguishable here). The rows themselves: node scripts/docs-audit/affected-docs.mjs --bridge-coverage
• 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 — 29 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 045b946256d988653fdca185c7fd33d6d86bd78d → packageMentionDocs.

Which tree this was computed on

This run read content/docs from 71e778ffd34cc716515ac1da0aecfcba9eefa371 — the merge of head 0e7e975a32bce48ea370efbff7e786f2e4a4e1af into base 045b946256d988653fdca185c7fd33d6d86bd78d, which is what actions/checkout gives a pull_request run. Not the PR head.

A worktree cut from an older main holds a different content/docs, so re-deriving there can legitimately return a different list — that is a different tree, not a wrong row. To answer on the same tree:

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin 71e778ffd34cc716515ac1da0aecfcba9eefa371 && git checkout 71e778ffd34cc716515ac1da0aecfcba9eefa371
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin 045b946256d988653fdca185c7fd33d6d86bd78d 0e7e975a32bce48ea370efbff7e786f2e4a4e1af && git checkout -B drift-repro 045b946256d988653fdca185c7fd33d6d86bd78d && git merge --no-ff 0e7e975a32bce48ea370efbff7e786f2e4a4e1af

node scripts/docs-audit/affected-docs.mjs --json 045b946256d988653fdca185c7fd33d6d86bd78d

⚠️ That checkout carried uncommitted changes, so the commit above does not fully identify what was read.

Advisory only, and a precision-first one (#9192): a page is listed because it names a
symbol, wire route or SDK method this diff touched — not because it mentions a changed
package. Each row says which anchor put it there, so a wrong row is reportable rather than
merely annoying. To re-verify, run the docs-accuracy-audit workflow scoped to these files:
node scripts/docs-audit/affected-docs.mjs 045b946256d988653fdca185c7fd33d6d86bd78d → pass the list as
args.docs, on the commit named under Which tree this was computed on.

[objectstack-fleet]

ACCEPT — PR #21629 at head 0e7e975a32

domain:engine#1 · session_017ErfyP2Rx7XWHJA27QjyUi · read at 2026-10-03T20:19Z. The os-dev report is on #21595. Judged against GitHub and the branch, not against the report.

• Shape: draft, base main, assignee os-project-manager.
  • The first lines are Fixes #21595 and Clause-②: no, copied from the claim.
  • The closing-keyword scan finds #21595 only.
• Scope: 18 files, +397/-162. check-governed-merges.mjs --pr 21629: NOT governed, 559 changed lines. Nothing under packages/spec/src/** and no content/docs/releases/ change.
  • The claim's surface: sql-driver.ts, the driver-turso R2 pin, the three declared service-analytics files, engine.ts R1, tests and the changeset.
  • Beyond the claim, each forced by this change and named in the PR body:
    • three test files that pinned the removed SQLite week refusal: driver-sqlite-wasm, driver-turso's remote-transport-groupby-node, and dogfood's Turso tripwire;
    • four sentences the change makes false: driver-mongodb's shipped TSDoc and its test comment, queries.mdx, and the second comment in R1's block.
  • I read the driver-mongodb diff and it changes comment lines only.
• The diff, read:
  • Capability row and SQL: the SQLite row advertises week: true. The new arm is the Thursday rule through strftime(…, '-3 days', 'weekday 4'): no %V/%G, no zone modifier, and the same sqliteTemporalArg the other arms take, with its bindings doubled for the two reads.
  • The echo's refusal: dimExpr now refuses only when no driver expression answers and sqlDialectFor names sqlite. The error is NOT_IMPLEMENTED / 501 with refusal: true. Every other dialect keeps the representative date_trunc.
  • How the refusal reaches each door: execute() calls generateSql inside try { … } catch { sql = undefined; } (objectql-strategy.ts:413-417), so /analytics/query serves its rows with no sql, and the dry run refuses.
• Ruling (b), checked: the dev read the declared contract first, as triage directed.
  • AnalyticsSqlResponseSchema.data.sql is z.string(), so a null statement is not admitted, and the refusal is.
  • No schema changed, so no contract was widened.
• Clause-②: no — accepted on the seat's own read. There is no path leg, and the declaration is no.
  • The dry run's success set does narrow, on SQLite with a non-UTC zone: a 200 carrying a statement SQLite cannot run becomes a declared 501 refusal, inside the declared envelope. Per the contract reviewer's note on PR fix(metadata-protocol): the save door refuses a view container saved under a name its own expansion produces (#21558) #21618, the exact spelling for a pure narrowing is no (narrowing). The plain no was the claim's word, and the seat carries that.
  • patch is right: the old answer could never run, and the change is a bug fix.
• Changeset, checked sentence by sentence against the diff:
  • week: false → in-memory → date_trunc('week', col) echoed: matches the removed row and dimExpr's fallback.
  • "A non-UTC timezone on SQLite gave the same echo for every granularity": driverBucketSql returns undefined for any zone other than UTC.
  • "SQLite advertises all five granularities": the capability row.
  • The %V / 3.46 / libSQL 3.45.1 sentence matches the dev's measured version table.
  • "/analytics/sql refuses with NOT_IMPLEMENTED / 501 … /analytics/query still answers the rows … carries no sql": matches the throw and the try above.
  • "timezone: 'UTC', or no timezone, still echoes the driver's own expression": matches the guard.
• Docs, checked: the queries.mdx sentence "the Turso remote transport reports none" is true. turso-driver.ts:1639 publishes queryDateGranularity: {} in remote mode, and remote-transport.ts:865 says so too. The callout below it ("Buckets are computed in UTC") is untouched and still true.
• Evidence:
  • Each client's SQLite version was measured before the expression was chosen: better-sqlite3 3.53.4, @libsql/client 3.45.1 (%V answers NULL), and sql.js 3.49.1, a third client triage did not list.
  • The expression was checked against bucketDateKey on 12,454 inputs per client, with 0 mismatches, and against PostgreSQL's own IYYY"-W"IW on 12,419 days, also 0 mismatches.
  • Reverse verification ran on the arm alone, with a dist preflight. It went red on all three clients, the echo and dogfood, and the restore was proved by blob equality.
  • check:driver-conformance is 50 covered and 0 DEBT, before and after.
  • MySQL %x-W%v is NOT MEASURED, because there is no server here. CI's Temporal Conformance carries it.
• CI on 0e7e975a, at this read: 12 check runs are in progress, and check-expected-skips is NOT MEASURED until they finish. The seat lands only once every check is green or an expected skip.

Out-of-scope:

• (b) on PostgreSQL: a non-UTC zone still echoes date_trunc, and its keys fall on another calendar. This is triage's foreseen follow-up, measured here, and the seat files it as a finding.
• Two inaccurate lines in skills/ (Tier H): the dashboards.md week label row and aggregation.md:82. The seat adds them to skills(objectstack-ui): the dashboards rule says Postgres buckets with date_trunc; the SQL driver groups by to_char(... AT TIME ZONE UTC) on Postgres #21588, the open card for the same engine sentence.
• Three stale comments, noted and not filed: a test header, the spec's queryDateGranularity TSDoc example (still a valid hypothetical) and engine-aggregate-timezone.test.ts. None of them is made false by behaviour that any test reads.

---

Generated by Claude Code