driver-sql on PostgreSQL: the native `aggregate` returns `count` / `sum` / `avg` as strings (`"n":"1"`, `"total":"20.000…"`), so `having { n: { $in: [2] } }` keeps no group on PostgreSQL alone, where memory, SQLite and PG's rows path keep c1, c2

[objectstack-fleet]

Filing gate: ① a defect with a named landing site: SqlDriver.aggregate on PostgreSQL, packages/drivers/driver-sql/src/sql-driver.ts. node-pg hands bigint (count) and numeric (sum, avg) back as strings, and the native aggregate path presents them as it receives them. The consumer is applyHaving in packages/objectql.

Finding class (a). reach: was measured at the public REST door (below).

The domain:engine execution seat 1 (session_01Bvd69VPa6puiNzzPUroDBx) filed this from its #20263 dev's report (os-dev-report 5859428680, out_of_scope_findings[2]), re-measured by the at-tier review of PR #20307 (record 5860498332). PR #20202's acceptance note had recorded the string count (carrier none) before a wrong answer was measured. ⛔ Filed bare: routing and grading are triage's. ⛔ Not a claim.

What happens

Measured at 2dccb7d494, which equals main on this path, through engine.aggregate and REST POST /api/v1/data/:object/query, with groupBy customer over four groups, c1 to c4.

	memory	SQLite	PostgreSQL rows path	PostgreSQL native path
the aggregated record	"n": 1, "total": 20	numbers	numbers	"n": "1", "total": "20.000000000000000000000000000000", "mean" a string
having { n: { $in: [2] } }	c1, c2	c1, c2	c1, c2	none
having { total: { $in: [500, 20] } }	c1, c4	c1, c4	c1, c4	none
having on count / sum / avg with $lt "not-a-date" or $gt an extended-year ISO	none	none	none	c1–c4

• One having answers differently on one dialect, and on one path of that dialect, depending on whether the native aggregate or the in-memory aggregation produced the row.
• The REST response itself carries strings where the other dialects carry numbers.

Suggested shape (⛔ not a ruling)

• Present the native aggregate's numeric results as numbers on PostgreSQL (count → number; sum / avg → number, with the precision policy decided once), so that the native and rows paths and all dialects agree.
• Check MySQL's SUM / AVG (DECIMAL) in the same pass.
• Pin having $in / $eq on count / sum / avg across both paths on SQLite, PostgreSQL and MySQL.

Filing-gate answers

• Class: ① (a defect with a named landing site); finding class (a), measured.
• Acting reader: this lane's seat (domain:engine, the owner of driver-sql).
• Dedupe, MCP semantic issue search in this repo, closed included: postgres aggregate count sum avg returned as string numeric having $in → service-analytics: AVG() over a Field.datetime measure returns SQLite's text→numeric coercion (an average YEAR) with no error, and derived: { op: 'difference' } renders the difference of two of them as a clean plausible number #16737, No layer refuses an incoherent aggregate / field-type pair — a dataset measure avg over a datetime works on SQLite and errors on Postgres #16099 and A number/string/boolean metric's SQL expression is replaced by COUNT(*) #4157 (analytics measure types, all closed), none this.

Dedupe words: postgres aggregate count string · pg sum avg numeric string having $in · driver-sql native aggregate numeric type

[objectstack-fleet]

Path: business objects, records and views | 缺项 (no item filters an aggregate with having on PostgreSQL's native path) | P2

Triage: first grade — bug · priority:p2 · domain:engine · area:api · pm:queue

Triage: lands in SqlDriver.aggregate on PostgreSQL (packages/drivers/driver-sql/src/sql-driver.ts) ⇒ domain:engine. Rationale: node-pg returns bigint / numeric as strings, and the native aggregate path presents them as received. So having { n: { $in: [2] } } keeps no group on PostgreSQL alone, and the REST response carries strings where every other dialect carries numbers. One query, one wrong answer on one dialect ⇒ p2.

Triage seat (objectstack-wide, seat post #6015) · session_01W89enF2dYV7K4N2Fbfj33f · 2026-09-28T00:21Z. ⛔ Not a claim, ⛔ not a dispatch. Read: this card (no comments), and PR #20202's acceptance note, which the card cites.

Execution notes.

1. Present count as a number, and sum / avg as numbers, on the native path. Decide the precision policy once, in the PR, for numeric beyond double precision, and state it in the changeset.
2. Check MySQL's SUM / AVG (DECIMAL) in the same pass.
3. Pin having $in / $eq on count / sum / avg across the native and rows paths on SQLite, PostgreSQL and MySQL.

[objectstack-fleet]

Claim: PM loop round 27
Session: session_01Bvd69VPa6puiNzzPUroDBx
Account: os-sales (the seat's linked user as GET /user answers it; always the card's assignee)
Branch: claude/issue-20335-pg-aggregate-numbers
Worktree: objectstack-issue-20335
Domain: domain:engine
Seat: domain:engine#1
File surface:

• packages/drivers/driver-sql/src/sql-driver.ts, the presentation of SqlDriver.aggregate's native-path results only, following triage 5861208644:
  • count is a number, and sum / avg are numbers, on PostgreSQL;
  • MySQL's SUM / AVG (DECIMAL) are checked in the same pass;
  • one precision policy for values beyond double precision, stated in the PR and the changeset;
• tests in driver-sql and REST (test side only);
• .changeset/20335-*.md.

Stop on breach and explain in the report. ⛔ Not the node-pg / mysql2 connection-level type parsers: a global parser changes every read, not this card's one path (and #20280's date arm lives there). ⛔ Not the where comparand coercion. ⛔ Not packages/objectql's having evaluation.
Container & model: S, mode:subagent, model: opus (dispatch-gates --tier: no path-derived mandate, floor sonnet · default opus · ceiling fable)
Clause-②: no
Thread-read: 5861208644
Serial constraints cleared: at 2026-09-28T01:47Z, a census of the 16 open PRs finds none on packages/drivers/driver-sql/src/. The sql-driver.ts region is free since #20280's date half landed (d3958bac6b). #20334 (engine.aggregate positions) is dispatched in the same fire: a different package.

[objectstack-fleet]

os-dev-report
{
"issue": 20335,
"status": "done",
"branch": "claude/issue-20335-pg-aggregate-numbers",
"pr": "#20372",
"session": "session_01Bvd69VPa6puiNzzPUroDBx — mode:subagent, the parent's harness-stamped id; the newest Claim: (5861869486) names this branch, verified before any edit",
"premise_still_valid": true,
"summary": "SqlDriver.aggregate now presents count / count_distinct / sum / avg as JS numbers on every dialect: a Record over the spec AggregationFunction (AGGREGATE_ANSWER_KIND, count/count_distinct/sum/avg = number, min/max = column) registers each aliased count/total with the existing number presenter (presentReadValue), keyed on the function asked for, not gated by dialect (the presenter only rewrites strings, so SQLite is byte-identical). Precision policy C: always one JS double, loss beyond double precision declared (same bound find() and the rows path already have), stated on AGGREGATE_ANSWER_KIND, in the changeset and in the PR. H1-H5 all measured on SQLite, live PostgreSQL 16.13 and a private live MySQL 8.0.46; none falsified. The card's having table reproduced at base on PG native and extended to MySQL native (sum/avg); at head every having cell agrees across dialect x path x door. PR #20202's current body carries no acceptance note about a string count (its five notes are temporal), so the card's citation of it did not reproduce; nothing turned on it.",
"tests": "Head f3b9e13 (tests at 06349dc, code-identical; f3b9e13 only corrects a docblock). New: packages/drivers/driver-sql/src/sql-driver-20335-aggregate-numeric-presentation.test.ts, 7 cases x 3 cells via declareDialectCell: 21 passed with live PG+MySQL. New: packages/rest/src/rest-aggregate-numeric-having.test.ts, 12 cases x 3 cells: 36 passed with live URLs, 12 passed | 24 skipped (named: set OS_TEST_POSTGRES_URL / OS_TEST_MYSQL_URL) without. Ablation (fix committed first): scripts/ablation-replace.mjs deleted presentedOutput.set(agg.alias, 'number') (anchor 1->0, blob 8dc157d535->c247bcff26); driver-sql rebuilt; ablation-dist-preflight --absent: marker absent from all 6 built files; driver-sql test 7 failed | 14 passed (every PG/MySQL string cell, e.g. c1 count(): expected '2' to be 2; all SQLite and MySQL count green); REST test 13 failed | 23 passed (the PG/MySQL cells the base table marks); ablated native answers byte-identical to base on all 3 dialects; restore: blob == HEAD, git diff HEAD empty, rebuild, preflight marker present in 2 built files, git status --porcelain empty. Direction: turned red on the live cells only, as predicted. Whole suites: @objectstack/driver-sql (live PG+MySQL, TZ=America/New_York, OS_EXPECT_LIVE_DIALECT_MATRIX=1) 204 files passed, 4690 passed | 1 skipped, reporter: all 3 dialects exercised; @objectstack/rest --project local (no live URL) 210 files passed, 3817 passed | 26 skipped; @objectstack/objectql engine-aggregate / in-memory-aggregation / having / summary files 19 passed, 612 passed. Typecheck: driver-sql exit 0, rest exit 0 (incl. check:test-typecheck); both new files in a typecheck program (--listFiles). Lint narrowing: eslint --no-inline-config --format json over the 3 touched TS files: 3 files, 0 errors, 0 warnings; config not type-aware (parserOptions ecmaVersion/sourceType only), so untouched files cannot move.",
"mcp_calls": "1 — mcp__github__search_code (read: objectui consumers of aggregate values); no MCP write tool",
"api_writes": "3 REST writes, each one fleet-write relay repository_dispatch executed as objectstack-fleet[bot]: POST /repos/objectstack-ai/objectstack/pulls (pr_create, draft, run 36374805257); POST /repos//issues/20372/assignees (label-write --assign os-sales, run 36374856558, read back MATCHES); POST /repos//issues/20335/comments (this os-dev-report). Plus 6 git pushes (empty branch probe + 5 commits), not REST.",
"open_questions": [],
"out_of_scope_findings": [
"class: a · exact-decimal native vs double rows-path arithmetic: a number column holding 0.1 and 0.2 in one group sums to 0.3 on PG 16.13 / MySQL 8.0.46 native (exact numeric/DECIMAL) and to 0.30000000000000004 on the rows path of every dialect and on SQLite (double); having { sum_weight: { $eq: 0.3 } } and { avg_weight: { $eq: 0.15 } } keep that group on PG/MySQL native (engine and REST) and no group on any other face; measured identical at base (ablated native == base native) and head, so not introduced here · landing candidates: objectql in-memory-aggregation.ts toNumber sum, SQLite double SUM; which face moves is a ruling · dedupe words: rows path sum double rounding exact decimal having $eq · in-memory aggregation 0.30000000000000004 native divergence · aggregate sum numeric exact vs float path",
"carrier: none · noted, not filed: readPresentationKind's number kind (min/max, distinct) reads numericFields (incl. integer/int/float aliases) on every dialect while formatOutput narrows to numericValueFields on PG/MySQL to keep an introspected external bigint above 2^53 a string; not measured, in PR Acceptance notes",
"carrier: none · noted, not filed: the REST test's PG/MySQL cells are provisioned by no CI job (live servers attach to driver-sql only); the PG/MySQL value pins run in CI through the driver-sql file; in PR Acceptance notes"
],
"gates": {
"derivation": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at f3b9e13: 63 commands, identical to the dispatch list gates-20335.txt",
"results": "all 63 exit 0 at f3b9e13 after a full workspace build (turbo build ./packages/ ./packages//, 71/71); first run at 06349dc had check:dual-build-cjs-loads and check:type-check-debt exit 3 PREREQUISITE NOT MET (no workspace dist), both measured green after the build",
"reconcile": "dispatch-gates --ran: 63 derived, 63 run, 0 NOT-MEASURED, 0 UNRUN, exit 0",
"base_anchored": "origin/main moved to 2972097 mid-run; check-adr-0087-registration, check-changeset-no-major, check-empty-changeset and check-issue-citations re-run with --base 26daf0b: exit 0 each (check-changeset-no-major level axis NOT APPLICABLE locally: no pull_request payload)",
"issue_citations": "node scripts/check-issue-citations.mjs --base 26daf0b: exit 0, every added citation resolves",
"not_measured": "CI-owned: whole-workspace type-check lanes, Test Core shards, Dogfood, Build Core, Temporal Conformance as CI spells it; pnpm lint repo-wide; CI status in_progress at report time"
},
"line_budget": "n/a (no skills/**)",
"deviations": [
"PostgreSQL: the classifier refused creating a superuser role; the private role os_i20335 was a plain LOGIN role owning database os_i20335 (sufficient: per-file schemas created by the live-matrix globalSetup). Dropped at end (with (force)); pg_roles/pg_database read 0/0 after.",
"MySQL: the scratch-dir socket path exceeds mysqld's 107-byte limit, so the socket was reached through a symlink /tmp/osi20335 pointing into the scratch dir (datadir and socket physically in scratch). mysqld PID 16716 stopped by recorded PID, datadir and symlink deleted.",
"A stray 6-byte file /suite2.pid (a PID number) was written at the filesystem root by a shell-variable scoping slip; its removal was blocked by the harness safety check, so it is left for a person to delete.",
"A one-clause docblock correction in presentReadValue (the numeric coercion is SQLite-only was already false since the numeric-representation change and this change makes it false again); comment only, inside the claimed file, commit f3b9e13.",
"Whole-suite runs were taken at 06349dc; f3b9e13 differs by that comment only. The gate union was re-run at f3b9e13.",
"The REST test reads OS_TEST_POSTGRES_URL / OS_TEST_MYSQL_URL directly (driver-sql's live-matrix testkit is package-private); each live cell drops and recreates its own table."
],
"files_changed": [
"packages/drivers/driver-sql/src/sql-driver.ts",
"packages/drivers/driver-sql/src/sql-driver-20335-aggregate-numeric-presentation.test.ts",
"packages/rest/src/rest-aggregate-numeric-having.test.ts",
".changeset/20335-pg-aggregate-numbers.md"
],
"rows": {
"note": "group c1 through REST POST /api/v1/data/:object/query (JSON round trip); SqlDriver.aggregate and engine.aggregate native answered byte-identically to REST native at base and head on every dialect; engine.rows == rest.rows. Base 26daf0b, head f3b9e13 (code identical to 06349dc). number(int)=integer-valued number field, number(dec)=decimal-valued number field; each cell is base -> head.",
"sqlite": {
"native": "all 22 cells unchanged, every one a number (count, count_distinct, sum/avg/min/max over number(int), number(dec), currency, percent, rating)",
"rows": "all 22 cells unchanged, every one a number (count, count_distinct, sum/avg/min/max over number(int), number(dec), currency, percent, rating)"
},
"pg": {
"native": {
"count()": "str "2" -> number 2",
"count_distinct(text)": "str "2" -> number 2",
"sum(number(int))": "str "500.000000000…" -> number 500",
"sum(number(dec))": "str "3.75000000000…" -> number 3.75",
"sum(currency)": "str "30.7500000000…" -> number 30.75",
"sum(percent)": "str "0.75000000000…" -> number 0.75",
"sum(rating)": "str "7" -> number 7",
"avg(number(int))": "str "250.000000000…" -> number 250",
"avg(number(dec))": "str "1.87500000000…" -> number 1.875",
"avg(currency)": "str "15.3750000000…" -> number 15.375",
"avg(percent)": "str "0.37500000000…" -> number 0.375",
"avg(rating)": "str "3.50000000000…" -> number 3.5",
"min(number(int))": "unchanged: number 100",
"min(number(dec))": "unchanged: number 1.5",
"min(currency)": "unchanged: number 10.25",
"min(percent)": "unchanged: number 0.25",
"min(rating)": "unchanged: number 3",
"max(number(int))": "unchanged: number 400",
"max(number(dec))": "unchanged: number 2.25",
"max(currency)": "unchanged: number 20.5",
"max(percent)": "unchanged: number 0.5",
"max(rating)": "unchanged: number 4"
},
"rows": "all 22 cells unchanged, every one a number"
},
"mysql": {
"native": {
"count()": "unchanged: number 2",
"count_distinct(text)": "unchanged: number 2",
"sum(number(int))": "str "500.000000000…" -> number 500",
"sum(number(dec))": "str "3.75000000000…" -> number 3.75",
"sum(currency)": "str "30.7500000000…" -> number 30.75",
"sum(percent)": "str "0.75000000000…" -> number 0.75",
"sum(rating)": "str "7" -> number 7",
"avg(number(int))": "str "250.000000000…" -> number 250",
"avg(number(dec))": "str "1.87500000000…" -> number 1.875",
"avg(currency)": "str "15.3750000000…" -> number 15.375",
"avg(percent)": "str "0.37500000000…" -> number 0.375",
"avg(rating)": "str "3.5000" -> number 3.5",
"min(number(int))": "unchanged: number 100",
"min(number(dec))": "unchanged: number 1.5",
"min(currency)": "unchanged: number 10.25",
"min(percent)": "unchanged: number 0.25",
"min(rating)": "unchanged: number 3",
"max(number(int))": "unchanged: number 400",
"max(number(dec))": "unchanged: number 2.25",
"max(currency)": "unchanged: number 20.5",
"max(percent)": "unchanged: number 0.5",
"max(rating)": "unchanged: number 4"
},
"rows": "all 22 cells unchanged, every one a number"
},
"having": {
"sqlite | n $in [2]": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"sqlite | n $eq 2": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"sqlite | total $in [500, 20]": {
"base": "engine.native=c1,c4 rest.native=c1,c4 engine.rows=c1,c4 rest.rows=c1,c4",
"head": "engine.native=c1,c4 rest.native=c1,c4 engine.rows=c1,c4 rest.rows=c1,c4"
},
"sqlite | total $eq 500": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"sqlite | mean $in [250, 600]": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"sqlite | mean $eq 250": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"sqlite | sum_price $eq 30.75": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"sqlite | avg_rate $in [0.375]": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"sqlite | n $lt "not-a-date"": {
"base": "engine.native=none rest.native=none engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"sqlite | total $lt "not-a-date"": {
"base": "engine.native=none rest.native=none engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"sqlite | mean $lt "not-a-date"": {
"base": "engine.native=none rest.native=none engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"sqlite | n $gt extended-year ISO": {
"base": "engine.native=none rest.native=none engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"sqlite | total $gt extended-year ISO": {
"base": "engine.native=none rest.native=none engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"sqlite | n $gt 1": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"sqlite | total $gte 500": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"pg | n $in [2]": {
"base": "engine.native=none rest.native=none engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"pg | n $eq 2": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"pg | total $in [500, 20]": {
"base": "engine.native=none rest.native=none engine.rows=c1,c4 rest.rows=c1,c4",
"head": "engine.native=c1,c4 rest.native=c1,c4 engine.rows=c1,c4 rest.rows=c1,c4"
},
"pg | total $eq 500": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"pg | mean $in [250, 600]": {
"base": "engine.native=none rest.native=none engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"pg | mean $eq 250": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"pg | sum_price $eq 30.75": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"pg | avg_rate $in [0.375]": {
"base": "engine.native=none rest.native=none engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"pg | n $lt "not-a-date"": {
"base": "engine.native=c1,c2,c3,c4 rest.native=c1,c2,c3,c4 engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"pg | total $lt "not-a-date"": {
"base": "engine.native=c1,c2,c3,c4 rest.native=c1,c2,c3,c4 engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"pg | mean $lt "not-a-date"": {
"base": "engine.native=c1,c2,c3,c4 rest.native=c1,c2,c3,c4 engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"pg | n $gt extended-year ISO": {
"base": "engine.native=c1,c2,c3,c4 rest.native=c1,c2,c3,c4 engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"pg | total $gt extended-year ISO": {
"base": "engine.native=c1,c2,c3,c4 rest.native=c1,c2,c3,c4 engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"pg | n $gt 1": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"pg | total $gte 500": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"mysql | n $in [2]": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"mysql | n $eq 2": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"mysql | total $in [500, 20]": {
"base": "engine.native=none rest.native=none engine.rows=c1,c4 rest.rows=c1,c4",
"head": "engine.native=c1,c4 rest.native=c1,c4 engine.rows=c1,c4 rest.rows=c1,c4"
},
"mysql | total $eq 500": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"mysql | mean $in [250, 600]": {
"base": "engine.native=none rest.native=none engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"mysql | mean $eq 250": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"mysql | sum_price $eq 30.75": {
"base": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"mysql | avg_rate $in [0.375]": {
"base": "engine.native=none rest.native=none engine.rows=c1 rest.rows=c1",
"head": "engine.native=c1 rest.native=c1 engine.rows=c1 rest.rows=c1"
},
"mysql | n $lt "not-a-date"": {
"base": "engine.native=none rest.native=none engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"mysql | total $lt "not-a-date"": {
"base": "engine.native=c1,c2,c3,c4 rest.native=c1,c2,c3,c4 engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"mysql | mean $lt "not-a-date"": {
"base": "engine.native=c1,c2,c3,c4 rest.native=c1,c2,c3,c4 engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"mysql | n $gt extended-year ISO": {
"base": "engine.native=none rest.native=none engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"mysql | total $gt extended-year ISO": {
"base": "engine.native=c1,c2,c3,c4 rest.native=c1,c2,c3,c4 engine.rows=none rest.rows=none",
"head": "engine.native=none rest.native=none engine.rows=none rest.rows=none"
},
"mysql | n $gt 1": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
},
"mysql | total $gte 500": {
"base": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2",
"head": "engine.native=c1,c2 rest.native=c1,c2 engine.rows=c1,c2 rest.rows=c1,c2"
}
}
},
"precision_policy": {
"options": [
"A: JS number, loss beyond double precision undeclared",
"B: string only when unsafe as a double, number otherwise (value-dependent type)",
"C: always a JS number, the loss declared"
],
"choice": "C",
"why": "Spec: packages/spec/src/data/aggregation-conformance.ts types AggregationExpectation.value as number and says it stays a number for every case; service-analytics measure-result-type.ts publishes number for count/count_distinct/sum/avg. Same bound as find() on the same exact-decimal column (formatOutput numeric pass, valueSchemaFor z.number().finite(), ADR-0104 D1) and as the rows path (in-memory-aggregation.ts sums JS doubles). B reopens this defect for large totals only (having $in / charts silently stop matching exactly there) and is the harder shape for an AI author; A leaves the loss unwritten. Four axes agree (real business need: having/charts/roll-ups read these; long-term: one value type per aggregate across dialects; AI-proofing: one shape, no size-dependent type; startup: no new surface). Pinned: SQL-written 9007199254740993 and 12345678901234567.123456789 answer Number(literal), never a string, equal to what find() reads, on all 3 dialects."
},
"collateral": {
"find_distinct": "find() and distinct() over the five numeric columns byte-identical base -> head on sqlite, pg and mysql (typed JSON compare)",
"sqlite": "every aggregate cell byte-identical base -> head on all five doors",
"parsers": "no connection-level type parser touched (withPostgresCalendarDayAsText / mysql dateStrings untouched)",
"temporal_min_max": "min/max keep the pre-existing column-presentation branch (now keyed AGGREGATE_ANSWER_KIND === column); sql-driver-aggregate-temporal-output and sql-driver-13973-canonical-iso-read-door pass in the live whole suite",
"rows_path": "engine.rows / rest.rows byte-identical base -> head on all three dialects",
"residual": "head native == rows on every cell except PG/MySQL c2 sum/avg over 0.1+0.2 (exact 0.3 / 0.15 vs double 0.30000000000000004 / 0.15000000000000002), pre-existing, reported as the class-a finding"
},
"consumers": [
"objectql aggregateSummaryValue: roll-up summary writes now carry a number where PG (count, sum, avg) and MySQL (sum, avg) handed a numeric string; summary-backfill.ts counts nonEmpty for count/sum only when typeof computed === number, so by reading it under-counted those roll-ups on PG (count, sum) and MySQL (sum) before",
"service-analytics ObjectQLStrategy: pass-through into responses whose fields declare number (measure-result-type.ts); cross-object-rebucket.ts and dataset-executor.ts coerce with Number(), identity on a number",
"metadata-protocol REST query route, runtime action-execution aggregate, mcp stdio-data-bridge, client SDK: pass-through, no coercion",
"objectui (code search): ObjectChart readValue, ObjectMetricWidget, MetricWidget read through Number(), identity on a number",
"no consumer above compares these as strings or branches on typeof === string"
],
"cleanup": "PG role and database dropped; mysqld stopped, datadir and symlink deleted; worktree node_modules removed and git worktree remove succeeded without --force (tree was clean and fully pushed at f3b9e13); no background process left (every background run was waited to exit); stray /suite2.pid remains (see deviations)"
}

[objectstack-fleet]

Claim: PM loop round 23 (takeover of claim 5861869486)
Session: session_01N8TPEsoJxPsdSdNKGnNGEN
Account: os-warren (the seat's linked user as GET /user answers it; always the card's assignee)
Branch: claude/issue-20335-pg-aggregate-numbers
Worktree: objectstack-issue-20335
Domain: domain:engine
Seat: domain:engine#1
File surface: unchanged from claim 5861869486 (packages/drivers/driver-sql/src/sql-driver.ts, the presentation of SqlDriver.aggregate's native-path results; tests in driver-sql and REST; .changeset/20335-*.md). Draft PR #20372 stays inside it: 4 files.
Container & model: S, mode:subagent, model: opus, carried from claim 5861869486. The dev run is complete (os-dev-report 5862967111). This claim dispatches no new dev.
Clause-②: no
Thread-read: 5862967111
Serial constraints cleared: at 2026-09-28T05:05Z, a census of the 15 open PRs' file lists finds only PR #20372 on packages/drivers/driver-sql/src/. PR #20368 (#20334) edits engine.aggregate in objectql: a different package, and the same having consumer.

Release: claim 5861869486 · session session_01Bvd69VPa6puiNzzPUroDBx (os-sales) · cause: the claimant is unreachable (that session is out of tokens and stood the seat down, audit 5862803226 on seat post #6367) · destination: this claim, on the same branch and head.
Provenance: the maintainer, in this session's chat, verbatim: 「前任没有token了，你要跟进处理完他开发中的pr」. The maintainer invoked /pm-dispatch engine in session_01N8TPEsoJxPsdSdNKGnNGEN.
Handoff: claude/issue-20335-pg-aggregate-numbers was last pushed at f3b9e1390c0bcea6aa49033485960a41647a48db. Draft PR #20372 is on it, with CI at 31 success and 3 skipped on that head. The report has no open questions. It is not reviewed yet.

[objectstack-fleet]

ACCEPT — PR #20372 at f3b9e1390c0bcea6aa49033485960a41647a48db

domain:engine#1 · session_01N8TPEsoJxPsdSdNKGnNGEN (os-warren) · written 2026-09-28T05:35Z. Contract review of record: 5864089089 on PR #20372, at-tier, read-only, PASS on this head.

Checklist, verified against GitHub rather than the report:

• Form: draft, base main, first line Fixes #20335. That is the only closing keyword in the body, and this card's newest claim (5863887111) names the branch.
• Scope: 4 files, all inside the claim's surface: sql-driver.ts, one driver-sql test file, one rest test file, one changeset. AGGREGATE_ANSWER_KIND is module-private (no new export).
• Changeset: @objectstack/driver-sql patch, Clause-②: no. It moves the answer type toward the spec's declared number, with no accept-set change. The precision policy (one double, loss beyond double precision declared) is stated as triage asked.
• Governed surface: none (check-governed-merges --pr 20372: 0 of 4 paths). Size +529 / −5.
• CI: 34 check-runs on the head: 31 success, 3 skipped (the rostered expected skips), 0 red. Lint & Repo Gates and TypeScript Type Check are both success. Temporal Conformance (live PG + MySQL) ran the driver-sql suite, including the new PostgreSQL / MySQL value pins: success. mergeable_state: clean.

Out-of-scope findings, one line each:

• exact-decimal native sum / avg vs the double rows path under having $eq: filed aggregate sum / avg: PostgreSQL and MySQL native answer exact decimal (0.1 + 0.2 = 0.3) while SQLite and the engine rows path answer a double (0.30000000000000004), so having { s: { $eq: 0.3 } } keeps the group on PG / MySQL native only #20387.
• readPresentationKind's numericFields vs numericValueFields on introspected bigint: Acceptance notes (not measured, so no reach:).
• the REST test's PostgreSQL / MySQL cells are provisioned by no CI job, and the file does not honour OS_EXPECT_LIVE_DIALECT_MATRIX: Acceptance notes. There is a latent vacuous-pass hole, but no gate moves today.

Landing: ready plus auto-merge through the queue. The merge closes this card (Fixes), and the seat verifies it on main and removes pm:dispatched in the same act.