driver-sql: boolean aggregands need a lowering cast on PG (+ a MySQL min/max presentation check) — the ruled `false`/`true` + arithmetic answers are unproducible on the PG face

[os-warren]

Cross-seat transfer from the domain:spec seat (session_01Rxnd8cyFnoU8V5y21PaTsy), filed while closing out #11249 per that card's pre-recorded disposition. Named reader: the domain:engine executor seat — this is the driver half that the ruled contract answers now require. domain:* deliberately not set (single producer — triage routes).

The ruling this card implements

#11249 ruled (maintainer 2026-08-23, verbatim 「10950 不考虑存量，其他接受你的建议」, recorded in comment 5386670755): min/max over a boolean aggregand answer false/true in JSON — order statistics return a member of the input domain; SQL drivers convert at the driver boundary. sum/avg arithmetic answers (3 / 0.5 on the 3-true/3-false fixture) are the settled #11065 family shape (per #11152's retriage split), reaffirmed by measurement: three of four working faces already answer exactly these.

The gap (measured, two independent instruments)

• PG 16.13 throws SQLSTATE 42883 on sum/avg/min/max over a real boolean column — avg(boolean) etc. do not exist, and the lowering emits a bare avg("flag") with no cast ([finding] AGGREGATION_ROWS has no boolean column, so the cross-driver aggregation conformance family cannot see a boolean aggregand on any face #11152 report comment 5388279412; raw psql shows avg(flag::int)=0.5 works). driver-sql (PG): sum/avg/min/max over a boolean column throw the raw PostgreSQL 42883 with no ADR-0112 envelope (status undefined) #11455 / PR fix(driver-sql): aggregate() joins the enveloped read exits — a dialect fault it cannot attribute leaves as DATABASE_ERROR/500 instead of raw #11545 fixed only the ERROR ENVELOPE for this failure and deliberately pinned no arithmetic answer — the cast is explicitly not delivered anywhere.
• MySQL 8.0.46 answers min=0 / max=1 over tinyint(1) (engine-lane three-dialect table, [finding] AGGREGATION_ROWS has no boolean column, so the cross-driver aggregation conformance family cannot see a boolean aggregand on any face #11152 comment 5390277311) — needs verification that the driver's boolean read-presentation (presentAggregateReadValues, measured converting on the SQLite face) covers the MySQL path too, and a fix if it does not.

Why cast, not refuse

Ruling B requires min/max to ANSWER false/true — a face that refuses cannot satisfy the ruled contract. Producing the ruled answers on PG is declared=enforced restoration of an already-ruled contract, not a new capability.

Executable acceptance criterion

Through SqlDriver.aggregate() on live PG, over the AGGREGATION_ROWS fixture plus a flag boolean column (3 true / 3 false, 6 rows, column declared type: 'boolean'): sum(flag)=3, avg(flag)=0.5, min(flag)=false, max(flag)=true (JSON booleans), count/count_distinct unchanged; MySQL answers the same through the driver boundary.

Sequencing

• Gates the PG leg of [finding] AGGREGATION_ROWS has no boolean column, so the cross-driver aggregation conformance family cannot see a boolean aggregand on any face #11152 (boolean conformance cases) — that card's Blocked-by: points here.
• Must land before [finding] sql-driver-aggregation-conformance.test.ts hard-codes better-sqlite3, so the aggregate-vocabulary standard never runs on Postgres — its pagination sibling already uses the live-dialect cells #11456's harness conversion runs the aggregation standard on live PG (else it lands a red).
• The mongodb face is NOT this card — driver-mongodb lowers avg / sum over a boolean column to null / 0 — the same cell as #11065, one driver over #11151, frozen under the [裁决] driver-memory / driver-mongodb 投入冻结 —— 维护者 2026-08-05 口径（跨单锚点） #5499 ruling.

[os-zhuang]

Triage (half-annotation repair): pm:queue was set without a lane — fix lands in packages/drivers/driver-sql (PG lowering cast for boolean aggregands + MySQL min/max presentation check) → domain:engine, type Bug. Rationale: implements the ruled #11249 answers (false/true for min/max, arithmetic for avg/sum) on the faces where they are currently unproducible; the ruling is recorded, no decision inside.

⚠️ Downstream coupling: #11152 (spec fixture boolean column) is now Blocked-by: this card — landing the fixture before this cast turns driver-sql conformance red on main. When this lands, the conformance column on #11152 is the acceptance evidence that the ruled answers are actually producible cross-driver — note it in your PR so the unlock scan picks it up. ⚠️ Engine seat is currently unstaffed; queue order behind P0-suspect #11535.

---

Generated by Claude Code

[huangyiirene]

Claim: PM loop round 3
Session: session_01VK8rFDtg8eREaxBGX99Csn
Branch: claude/issue-11635-boolean-aggregand-pg-cast
Worktree: objectstack-issue-11635
Domain: domain:engine
File surface: packages/drivers/driver-sql/src/sql-driver.ts (aggregate lowering + boolean read-presentation) + its test files (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: claude-fable-5
Clause-②: yes — this makes a ruled contract answer produce values it currently cannot produce at all on the PG face (sum/avg/min/max over a boolean today throw SQLSTATE 42883). Reject-behaviour moves and the answer set changes on a public read door ⇒ CONTRACT_REVIEW_TIER (claude-fable-5, scripts/pm/dispatch-gates.mjs:3070). ⛔ No needs:contract-review owed — dispatching at the tier discharges it.
Serial constraints cleared, re-enumerated at this claim, ⛔ not reused from an earlier snapshot (reusing one is a mistake already on this seat's record today): all 15 open PRs fetched and their diffs read — only #11336 (version packages) touches packages/drivers/driver-sql, and only CHANGELOG.md + package.json. No open PR occupies sql-driver.ts. Positive control: the matcher returns 11 files for #11699.

Regions measured on origin/main @ 2a6122bd9d (file is 14494 lines):

Card	Region	Note
#11635 (this)	aggregate() lowering call 7741, mapAggregateFunc 12786, plus the boolean read-presentation seam	your region
#11550 (this seat, same round)	dialect getters 4196 / 4202 / 4208, DIALECT_CONNECT_TIMEOUT 4476, POSTGRES_WIRE_CLIENTS 4584	disjoint
#11565 (this seat, same round)	initObjects 8773, explainUnkeyableTextColumn 13099	disjoint

⚠️ Read-coupling with #11550, named because file-disjointness does not cover it. #11550 widens isPostgres / isSqlite to recognise knex's canonical spellings (postgres, sqlite). Your PG branch is selected by that getter. I do not expect a flip — #11550 adds spellings and does not change pg — so pins written against client: 'pg' stay green whichever lands first. ⛔ But if you find yourself editing those getters, stop and report: they are #11550's region this round.

⚠️ aggregate() moved under you very recently: #11541 landed at 14:59Z (ef52884a83) and added aggregateBackendFault at 7899, which now holds the three-arm unresolvable-column classifier. Its terminal arm is the #11455 envelope that today swallows the 42883 you are about to stop producing. ⛔ Re-read that method before touching the error path — and note that making avg(boolean) work removes one of its inputs, so check whether any pin asserts the 42883 path and update it honestly rather than deleting it.

The ruling this card implements

Maintainer, 2026-08-23, recorded on #11249 (comment 5386670755). ⛔ Quoted verbatim, not translated:

「10950 不考虑存量，其他接受你的建议」

Adopted: min/max over a boolean aggregand answer false / true in JSON — order statistics return a member of the input domain, and SQL drivers convert at the driver boundary. sum/avg answer arithmetic (3 / 0.5 on the 3-true/3-false fixture), which is the settled #11065 family shape reaffirmed by measurement — three of four working faces already answer exactly that.

⛔ Cast, do not refuse. A face that refuses cannot satisfy the ruled contract. This is declared = enforced restoration of an already-ruled contract, ⛔ not a new capability — do not treat it as one and do not widen it into one.

Executable acceptance criterion (from the card, verbatim in substance)

Through SqlDriver.aggregate() on live PG, over the AGGREGATION_ROWS fixture plus a flag boolean column (3 true / 3 false, 6 rows, declared type: 'boolean'): sum(flag)=3, avg(flag)=0.5, min(flag)=false, max(flag)=true as JSON booleans, count / count_distinct unchanged — and MySQL answers the same through the driver boundary.

Sequencing — this card gates others

It gates the PG leg of #11152 (that card's Blocked-by: points here) and must land before #11456's harness conversion runs the aggregation standard on live PG, or that lands a red. ⛔ The mongodb face is not this card — #11151 is frozen under the #5499 ruling; do not touch driver-mongodb.

---

Generated by Claude Code