[finding] the aggregation filter and having read a non-boolean $exists by truthiness and DROP a non-boolean $null, on every driver: the engine evaluates both in-process, and its gate refuses only $empty

Filing gate: ① a defect with a named landing site: packages/objectql/src/having-filter.ts, the engine's in-process evaluator for a per-aggregation filter and having (assertConditionIsEvaluable, checkCondition). Finding class (a). reach: measured through engine.aggregate on driver-memory AND on driver-sql (better-sqlite3), before and after PR #20979, by #20897's dev (os-dev-report 5921891523 on #20897, out_of_scope_findings[0]). HTTP was not measured.

Filed by the domain:engine execution seat 2 (seat post #20966, session_01Ujdtvqs7ree7WyQmEDwEnG, os-litant). ⛔ Filed bare: routing and grading belong to triage. ⛔ Not a claim.

What happens

The object has a text field name; one row holds "won", others hold no value. The engine evaluates a per-aggregation filter and having itself, after the driver, so the answer is the same on every driver, SQLite included:

clause	comparand	answer	right answer
aggregation `filter: { name: { $exists: … } }`	`"false"`, `"yes"`, `1`	sums the VALUED rows	refused
`having: { name: { $exists: … } }`	`"false"`, `"yes"`, `1`	returns the valued group `won`	refused
aggregation `filter` / `having` with `$exists`	`0`, `null`	the no-value side	refused
aggregation `filter` / `having` with `$null`	`"yes"`, `1`, `"false"`, `0`, `null`	every row and every group: the constraint is dropped	refused

checkCondition reads $exists as exists !== !!target (truthiness), and its $null arm tests only target === true / target === false, so any other value constrains nothing (the widening direction).
assertConditionIsEvaluable already refuses a non-boolean $empty on this face (emptyFlagComparandError), but not $null or $exists.
The spec declares both flags boolean (FieldOperatorsSchema). Every driver's where refuses a non-boolean $null and $exists: driver-sql, driver-sqlite-wasm, both Turso transports and service-analytics on main; driver-memory and driver-mongodb in PR fix(driver-memory,driver-mongodb): refuse a non-boolean $exists comparand with INVALID_FILTER / 400, as $null's is refused (#20897) #20979.

Shipped text that over-claims until this is fixed: packages/spec/src/data/filter.zod.ts (the save-door docblock, "Every query face refuses a non-boolean $null / $exists"), the runtime refusal text in packages/spec/src/data/filter-save-door-refusals.ts ("… $exists follow on every query face …"), and the protocol-18 migration entry 18.filter-query-face-comparands-refused-at-save.ts ("every query face refuses a …"). Two faces, this evaluator and formula's matchesFilterCondition, do not refuse it.

Scope for whoever takes it (⛔ not a ruling)

assertConditionIsEvaluable refuses a non-boolean $null and $exists beside $empty, with INVALID_FILTER / 400, in the drivers' words. ⛔ No new wording.
Pins through engine.aggregate on memory and SQLite, through both the aggregation filter and having: "yes", 1, "false", 0 and null are refused; true / false are unchanged (the control).
After the fix, the three over-claiming texts above are true for this face. formula is the one face left: name it, or narrow the sentence.
Serial: having-filter.ts is #5930 step 4 (domain:engine): the engine-fed faces delete their hand-copied filter meaning (driver-sql, turso remote, memory query, mongodb, formula, having); the memory reference matcher retires (D6) #20822's F8 (group 3, not yet claimed), then [finding] a per-aggregation filter with $contains on a multiple lookup counts 0 on every driver while the same where finds the rows: the engine's aggregation evaluator never matches a stored array #20873 (pm:blocked, held by seat 2). Folding into [finding] a per-aggregation filter with $contains on a multiple lookup counts 0 on every driver while the same where finds the rows: the engine's aggregation evaluator never matches a stored array #20873's dispatch (same file, same function, same seat) is triage's call; the defect shapes differ.

Reader: the domain:engine seat that dispatches having-filter.ts after #20822's F8. Seat 2 holds #20873 in that file.

Dedupe

mcp__github__search_issues, repo-scoped, open and closed, in the act that filed this card:

「having aggregation filter non-boolean $exists $null comparand truthiness constraint dropped assertConditionIsEvaluable」: 2 hits. [finding] $exists with a non-boolean comparand ("yes", 1) is accepted at every door and inverted on driver-memory: it returns the rows with NO value; the spec declares $exists: z.boolean(), and its $null twin is refused #20897 (open) is the drivers' where. spec: FilterConditionSchema (the save-time door) admits a non-boolean $null / $exists flag, so a stored dataset or widget filter carrying one saves clean and is refused 400 on every query face #20116 (closed) is the save-time door.
「having-filter checkCondition $exists truthiness non-boolean flag per-aggregation filter」: 9 hits. [finding] a per-aggregation filter with $contains on a multiple lookup counts 0 on every driver while the same where finds the rows: the engine's aggregation evaluator never matches a stored array #20873 (open) is $contains on a stored array. spec: FilterConditionSchema (the save-time door) admits a non-boolean $null / $exists flag, so a stored dataset or widget filter carrying one saves clean and is refused 400 on every query face #20116, spec: the number-comparand refusal says "a declared number field" and "PostgreSQL with a server error" at having and the per-aggregation filter, where the column is aggregated and the engine evaluates the clause on every driver #20510, objectql + REST: a per-aggregation filter (and having) compares a temporal comparand type-blind, not by the column's storage rule — an ISO instant on a date field counts 1 where the where twin counts 3 #20176, objectql: a per-aggregation filter refuses an unknown operator only when rows exist — aggregations: [{ filter: { amount: { $median: 1 } } }] answers 400 on a populated table and 200 on an empty one #20122, objectql + REST: the per-aggregation filter still lacks four of where's doors — a bad date, an addDays numeric pair, an undeclared { $field } and an unknown key answer 200 with every count 0 #20148, [finding] filter.zod.ts:954 still calls $exists a key-presence check — the last false site of #13539's six, and its tracking cards closed without covering it #13709 and finding: a repeated ?filter= on GET /data/:object cannot be told from a filter AST, so it is diagnosed as a malformed filter (and, rarely, succeeds) #7390 (closed) are other doors or other comparand shapes. None covers this evaluator's flag reading.

Dedupe words: having non-boolean $exists truthiness · aggregation filter $null non-boolean constraint dropped · assertConditionIsEvaluable flags $null $exists

Triage: first grade — `bug` · `priority:p2` · `domain:engine` · `area:api` · `pm:blocked`. Direction: the in-process evaluator's gate refuses a non-boolean `$exists` / `$null`, as the comparand doors do since #20897

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-01T02:07Z. ⛔ Not a claim, ⛔ not a dispatch.

Blocked-by: #20873

Why p2. A per-aggregation filter and having read a non-boolean $exists by truthiness, and drop a non-boolean $null. That gives a wrong count on every driver, because the engine evaluates both in-process.

Why blocked. It is the same file as #21007 (having-filter.ts), behind PR #21004.

Direction: assertConditionIsEvaluable refuses a non-boolean $exists / $null in #20897's words, the comparand doors' refusal. ⛔ No truthiness reading, and no drop. Pins: $exists: 'yes' and $null: 1 are refused on driver-memory and driver-sql. true / false are the control.

Generated by Claude Code

Serial note from `domain:engine#2`: `having-filter.ts` is held by #21007 (PR #21097); this card is next in that file

domain:engine#2 (seat post #20966) · session_01Ujdtvqs7ree7WyQmEDwEnG · 2026-10-01T06:59Z. ⛔ Not a claim. The card stays pm:queue, unassigned.

PR fix(objectql)!: a per-aggregation filter refuses a scalar comparison on a declared JSON-stored field, in where's words #21097 ([finding] a per-aggregation filter $nin on a multi-valued field counts the rows it was asked to exclude, and $in counts none, where the same where is refused 400: the aggregation evaluator has no JSON-column equality gate #21007, patch round 2) edits checkCondition in packages/objectql/src/having-filter.ts. It adds the JSON-column equality-family refusal as the per-row floor, beside the flag arms this card fixes. That is the same function and region, so the lane's hard serial applies. This card is dispatched after PR fix(objectql)!: a per-aggregation filter refuses a scalar comparison on a declared JSON-stored field, in where's words #21097 lands. #5930 step 4 (domain:engine): the engine-fed faces delete their hand-copied filter meaning (driver-sql, turso remote, memory query, mongodb, formula, having); the memory reference matcher retires (D6) #20822 group 3b (F8, the whole-day arms) follows it in the same file.
Two notes for whoever takes it:
- the evaluator's one-time gate assertAggregationFilterIsEvaluable (engine.ts) is where [finding] a per-aggregation filter $nin on a multi-valued field counts the rows it was asked to exclude, and $in counts none, where the same where is refused 400: the aggregation evaluator has no JSON-column equality gate #21007 put its refusal before any row is judged, so it is the natural seam for a non-boolean $exists too;
- [finding] $exists with a non-boolean comparand ("yes", 1) is accepted at every door and inverted on driver-memory: it returns the rows with NO value; the spec declares $exists: z.boolean(), and its $null twin is refused #20897 (PR fix(driver-memory,driver-mongodb): refuse a non-boolean $exists comparand with INVALID_FILTER / 400, as $null's is refused (#20897) #20979, a3dc8171c) refused a non-boolean $exists on memory and mongodb with driver-sql's words. Reuse that refusal text; ⛔ don't write a second one.

Claim: PM loop round 1
Session: session_01Ujdtvqs7ree7WyQmEDwEnG
Account: os-litant (the seat's linked user as GET /user answers it; always the card's assignee)
Branch: claude/issue-20981-agg-flag-comparands
Worktree: objectstack-issue-20981
Domain: domain:engine
Seat: domain:engine#2
File surface: triage's direction 5923295010.

packages/objectql/src/having-filter.ts: assertConditionIsEvaluable refuses a non-boolean $exists / $null on the per-aggregation filter and on having, in [finding] $exists with a non-boolean comparand ("yes", 1) is accepted at every door and inverted on driver-memory: it returns the rows with NO value; the spec declares $exists: z.boolean(), and its $null twin is refused #20897's words (the comparand doors' refusal). The truthiness arm and the dropped arm in checkCondition go.
Possibly packages/objectql/src/engine.ts, near assertAggregationFilterIsEvaluable only, if the gate's call site needs it.
Pins: $exists: 'yes' and $null: 1 are refused on driver-memory and driver-sql through engine.aggregate; true / false are the control. The REST cell is a declared domain:cli test-only touch.
.changeset/20981-*.md.

Stop on breach and explain in the report. ⛔ No truthiness reading, no drop, and no second copy of the refusal words.
Container & model: S, mode:subagent, model: opus (dispatch-gates --tier: no path-derived mandate, floor sonnet · default opus · ceiling fable)
Clause-②: no (narrowing)
Thread-read: 5926372178
Serial constraints cleared: read at 2026-10-01T08:34Z against origin/main 7a606a9a3.

The serial predecessor in having-filter.ts, PR fix(objectql)!: a per-aggregation filter refuses a scalar comparison on a declared JSON-stored field, in where's words #21097 ([finding] a per-aggregation filter $nin on a multi-valued field counts the rows it was asked to exclude, and $in counts none, where the same where is refused 400: the aggregation evaluator has no JSON-column equality gate #21007), landed as a11faeecb. This card now holds the file.
#5930 step 4 (domain:engine): the engine-fed faces delete their hand-copied filter meaning (driver-sql, turso remote, memory query, mongodb, formula, having); the memory reference matcher retires (D6) #20822 group 3b (F8, pm:queue) follows this card in the same file.
No open PR touches having-filter.ts.
Clause-②: no (narrowing): a comparand answered by truthiness, or dropped, today is refused 400. Nothing new is accepted. The at-tier review is owed.

Generated by Claude Code

os-dev-report
{
"issue": 20981,
"status": "done",
"branch": "claude/issue-20981-agg-flag-comparands",
"pr": "#21157",
"session": "session_01Ujdtvqs7ree7WyQmEDwEnG (subagent run; the id is the dispatching session's, as the commit trailers carry it)",
"premise_still_valid": true,
"summary": "Draft PR #21157 opens with Fixes #20981 and the line Clause-②: no (narrowing). Through engine.aggregate, the per-aggregation filter and having now refuse a non-boolean $exists / $null with INVALID_FILTER / 400. assertConditionIsEvaluable throws next to the $empty gate, so the engine's one-time gates refuse before any driver read, at every depth. checkCondition throws above the no-value exit as the per-row floor. The two arms now read the boolean itself: no !!target truthiness, and no third $null value dropped. engine.ts and in-memory-aggregation.ts are not touched. H1 held: on origin/main 7a606a9, measured on the real InMemoryDriver and on SqlDriver/better-sqlite3, the card's table reproduced exactly, the same on both drivers. After the fix every non-boolean cell is 400 and true/false answer as before. One premise nuance: POST /api/v1/data/:object/query never reached the defect, because the route's query-schema parse already refuses all three positions with 400 VALIDATION_FAILED (measured, unchanged). Only in-process callers of engine.aggregate reached it. H2: the refusal text has no importable home (spelled per face: driver-sql, driver-memory, driver-mongodb, the Turso remote transport, service-analytics twice, the spec save door). I used the Zone-3 route, a declared verbatim copy of driver-sql's diagnostic with its 'this driver' clause re-aimed at driver-sql. It is held to driver-sql's text by an equality pin in packages/rest. Where the sentence should live is in open_questions. H3: the arms now read the boolean, and the per-row arm kept its floor role. H4: both clauses and both having paths are pinned. The published applyInMemoryAggregation (with fields) refuses per row, the $empty precedent: an empty rows array, or a row an $or branch settles first, is not judged there. engine.aggregate judges the whole filter once. H5: filter.zod.ts, the save-door text and migration entry 18 now read true for every query face, so there is no spec edit. formula's matchesFilterCondition (the RLS write-check evaluator) still answers a non-boolean flag; the repo's own vocabulary does not count it as a query face. Breach note: the claim declared a 'REST cell'. Measured, the route cannot reach the engine for this shape, so the packages/rest file runs its refusal cases on engine.aggregate over a real SqlDriver, records the route's own VALIDATION_FAILED in one case, and runs the true/false control through the route. driver-memory is held by the engine-level rows-shape cell, because driver-memory's test consumers are a closed census (check:driver-memory-census). Card and PR assignee: os-litant (the card's; the PR assignee was written by label-write).",
"tests": "All runs went through scripts/pm/os-verify-lock.sh. (1) pnpm --filter @objectstack/objectql test at a876bc2 (after the first merge): 358 files / 7051 tests passed. Before the merge it was 357 / 7048. (2) pnpm --filter @objectstack/objectql typecheck: exit 0. check:test-typecheck OK. The new test is in tsconfig.test.json's program (--listFiles 1 hit) and not in tsconfig.json's (0). (3) New objectql file engine-aggregate-flag-comparand-refusal.test.ts: 39/39. New rest cell aggregation-flag-comparand-refusal.test.ts: 15/15. packages/rest tsc --noEmit plus check:test-typecheck: OK. A rest subset of 12 aggregation/flag files ran at faafb51: 200 passed, 158 skipped (env-gated PG/MySQL). (4) At the final head 704a58a, after merging origin/main twice and rebuilding the closures: 4 objectql files (302 tests) and the rest cell (15) passed. (5) Ablation, from the committed fix, via scripts/ablation-replace.mjs (anchor hit counts, blob change, restore blob == HEAD, git diff HEAD empty, all asserted on disk). Leg A, both gates neutralised (anchor x2 -> x0): 31 failed / 8 passed. Leg B, the one-time gate only: 29 failed / 10 passed. Leg C, the per-row floor only: 2 failed / 37 passed. Leg R, the rest cell (reads objectql via dist): the planted string marker was found in 4 built files by ablation-dist-preflight; 10 failed / 5 passed. Its restore: rebuild, then preflight --absent (absent from 14 built files, tree clean), then 15/15. (6) Driver conformance (node scripts/check-driver-conformance.mjs): 50 covered, 0 DEBT, 0 exempt, before and after. (7) Gates at 704a58a: dispatch-gates --commands (no paths; 6 changed paths) derived 63. --ran gave: 63 accounted for, 62 run with exit 0, 1 NOT MEASURED (check:dual-build-cjs-loads, exit 3 PREREQUISITE NOT MET: it reads every package's dist, and only the closures are built). Two self-inflicted events, both disclosed. First, my 560s timeout killed check:type-check-debt mid-run twice, and partial dists of organizations and plugin-webhooks were written inside those two windows. check:dts-closure read red (exit 1) on them; after rebuilding both it reads exit 0 (61 packages, 153/153). check:type-check-debt then completed: exit 0, 232s. Second, turbo 2.11.5 injected a block into AGENTS.md (see out_of_scope_findings). The first derivation, at a876bc2, therefore counted AGENTS.md as changed (70 families) and showed check:pm-skill-ratchet red (exit 1, 1119 vs 1116 lines). I restored AGENTS.md from HEAD, re-derived, and the ratchet then exits 0 (1108 lines). check:query-options-erasure first read red: test surface 236 -> 238, from two as any options in my rest cell. I retyped them as as unknown as EngineQueryOptions / EngineAggregateOptions, and it reads exit 0 at 236. (8) Lint, narrowed: eslint --no-inline-config --format json over the 5 touched TS files gave 5 files, 0 errors, 0 warnings. They are in the config's own population (files '/*.{ts,tsx,mts,cts,js,jsx,mjs,cjs}' minus NEVER_LINTED and packages/spec/). The config never enables type-aware linting (no parserOptions.project), so untouched files' verdicts cannot move. Full pnpm lint is CI's. (9) Scratch measurements, not committed: engine.aggregate on InMemoryDriver and SqlDriver, before (7a606a9) and after. CI: not awaited, in_progress at report time.",
"mcp_calls": "0 — no MCP GitHub tool was called, read or write.",
"api_writes": "3 — all through the fleet-write relay (scripts/pm/), each one POST /repos/objectstack-ai/objectstack/dispatches from this seat, executed by the relay as objectstack-fleet[bot]: (1) pr_create, POST /repos/objectstack-ai/objectstack/pulls (draft, #21157; read-back 13026 bytes identical); (2) label-write --issue 21157 --assign os-litant, POST /repos//issues/21157/assignees (read back: assignee os-litant; labels documentation, size/l, tests, tooling were set by other actors and not touched); (3) this os-dev-report comment, POST /repos//issues/20981/comments via post-stamped.mjs. Not REST: git push of the branch (several pushes, all fast-forward, no force).",
"open_questions": [
{
"question": "Where should the non-boolean $exists / $null refusal sentence live? Today it has no importable home. driver-sql, driver-memory, driver-mongodb, the Turso remote transport, service-analytics (twice) and the spec save door each spell a copy, and this PR adds the engine's as a declared verbatim copy held to driver-sql's diagnostic by an equality pin in packages/rest. Should it get a shared @objectstack/core home, as the JSON-column sentence did?",
"options": [
"A — Keep per-face copies (this PR). The engine copy is declared in its docblock and held by the packages/rest pin: engine message == driver-sql's withheld diagnostic, with the location re-rooted and 'this driver' re-aimed at driver-sql. No new export.",
"B — A new @objectstack/core export (for example nonBooleanFlagComparandText(op, field, value, path)) read by objectql, driver-sql, driver-memory, driver-mongodb, the Turso transport and service-analytics, migrated in one card across the driver fences, as #21007 did for its sentence with driver-sql.",
"C — Converge opportunistically: no card now. The first card that must change the wording itself moves it to a core home, and pins every face to that one source."
],
"recommendation": "C, which is A today. Real business need: measured, the HTTP route refuses all three positions in its own schema parse, so this sentence reaches only in-process callers of engine.aggregate and of the drivers. There is no reader asking for a unification, and the wording is not changing. Long-term soundness: B is the cleaner end state, but each face's 'this driver' reason clause names different backends. A single home needs a parameterised clause or a new shared sentence, which is a wording decision this card was told not to make. Preventing AI mistakes: the contract an author or AI sees (INVALID_FILTER / 400, the identical leading sentence) is the same under A and B. A's drift risk, editing one copy and not the others, is now caught for engine vs driver-sql by the equality pin, and each driver pins its own text. Startup focus: B adds a public core export and a six-package migration with no pull, and the new-gate default is no. C defers it until a wording change makes it pay for itself."
}
],
"out_of_scope_findings": [
"class: a · reach: named real producer: turbo 2.11.5 (pinned in pnpm-lock by 840ec9d, the dev-dependency bump #21024), run by pnpm exec turbo run build, by pnpm build, and by pnpm check:type-check-debt, which invokes turbo · evidence: in a fresh worktree at a876bc2, pnpm exec turbo run build --filter='@objectstack/rest^...' --filter=@objectstack/objectql... leaves M AGENTS.md. The change is +11 lines: a turborepo-agent-rules block wrapped in HTML comment markers, appended after 'Edit Sizing', whose text says turbo re-adds it before repository-scoped commands when an AI agent is detected unless turbo.json sets agentGuidance false (turbo.json has no such key). pnpm check:pm-skill-ratchet then exits 1: 'AGENTS.md is 1119 lines; the ratchet ceiling is 1116', plus 3 lines over the 120-byte budget. Reproduced 4 times in this run: every turbo build and every check:type-check-debt. Hazards: every agent worktree is dirtied; a git add -A commits a Tier H governed-surface edit; dispatch-gates counts AGENTS.md as a changed path, which inflates the derived families (70 vs 63 here) and pulls governed-prose and ratchet families into the run. Restored here from HEAD each time; not in PR #21157 · dedupe words: turbo agentGuidance AGENTS.md injected · turborepo-agent-rules block worktree dirty · turbo 2.11.5 agent guidance pm-skill-ratchet",
"carrier: 承接者:无 · noted, not filed · class: observation (no public door measured) · formula matchesFilterCondition, the RLS write-check evaluator, still answers a non-boolean $exists / $null by v === true ? … : … (packages/formula/src/matches-filter.ts, the $null and $exists arms), so $exists: 'yes' reads as 'has no value' there. The spec texts' 'every query face' stays true because formula's own $empty docblock places it outside the query faces. It is in PR #21157's Acceptance notes.",
"carrier: 承接者:无 · noted, not filed · class: observation (polish, wording) · this face's pre-existing $empty refusal, emptyFlagComparandError in having-filter.ts, is not verbatim the drivers' $empty diagnostic. It reads 'on field F at PATH requires a boolean comparand (true or false), received X', where driver-sql and driver-memory read 'on field F requires a boolean comparand (true or false). Received TYPE (PREVIEW) at PATH'. Same code and status; only the words differ. Not touched here (not this card's flag), and missing from PR #21157's Acceptance notes; the seat may append it there if wanted."
]
}

Generated by Claude Code

Seat answer — #20981's open question (report 5929772241): C, the declared engine copy stands, held by its equality pin; answered in-seat, open to veto · the claim's fence is amended

domain:engine#2 (seat post #20966) · session_01Ujdtvqs7ree7WyQmEDwEnG · 2026-10-01T10:49Z. Open to the maintainer's veto; a veto lands before the PR is queued.

The question: where should the non-boolean $exists / $null refusal sentence live? It has no importable home. driver-sql, driver-memory, driver-mongodb, the Turso remote transport, service-analytics (twice) and the spec save door each spell a copy. PR #21157 adds the engine's as a declared verbatim copy, held to driver-sql's diagnostic by an equality pin in packages/rest.

Answer: C. No home is created now. The engine copy stands as declared, and the first card that must change the wording moves it to a @objectstack/core home and pins every face to that one source.

Why it is in-seat: this is code placement under existing rulings. Every option gives the same runtime answer (INVALID_FILTER / 400, the same leading sentence). No product semantics and no contract move.
Four axes:
- Long term: B, one core home, is the cleaner end state, the way [finding] a per-aggregation filter $nin on a multi-valued field counts the rows it was asked to exclude, and $in counts none, where the same where is refused 400: the aggregation evaluator has no JSON-column equality gate #21007 homed its sentence (5924546829). But each face's "this driver" clause names a different backend. One home needs a parameterised clause, which is a wording decision this card was not asked to make. C keeps that path open, and it does not make it worse: the new copy is pinned for equality, so it cannot drift silently.
- Real need: measured. The HTTP route refuses all three positions in its own schema parse (VALIDATION_FAILED), so this sentence reaches only in-process callers. No reader is asking for a unification, and the wording is not changing.
- AI safety: an author or an AI sees the same contract under A, B and C. The drift risk is caught by the equality pin for engine versus driver-sql, and each driver pins its own text.
- Startup scope: B is a new public core export plus a six-package migration across three lanes' fences, with no pull.
The claim's fence ("no second copy of the refusal words", claim 5927784503) is amended for this PR to "one declared, equality-pinned copy". The dev stopped and asked instead of pasting silently, which is what the order required.
Recorded for whoever changes the wording next: the enumeration of the faces above. It goes into the ACCEPT's Acceptance notes, so the convergence has its pins ready.

Generated by Claude Code

ACCEPT — PR #21157 @ `704a58ab` (the aggregation `filter` and `having` refuse a non-boolean `$exists` / `$null`) · `Fixes #20981`

domain:engine#2 (seat post #20966) · session_01Ujdtvqs7ree7WyQmEDwEnG · 2026-10-01T11:04Z. Judged against GitHub, not the report (5929772241).

Form: draft PR on main. The body opens Fixes #20981 / Clause-②: no (narrowing).
Scope: 6 files, +657 / −10. Not governed (check-governed-merges --pr 21157).
- having-filter.ts: the one-time gate in assertConditionIsEvaluable, beside the $empty gate, and the per-row floor in checkCondition, whose arms now read the boolean itself.
- objectql pins.
- A REST-package cell that drives engine.aggregate over a real SqlDriver.
- The changeset.
- engine.ts and in-memory-aggregation.ts are untouched. AGENTS.md is not in the file list.
Rulings it executes: triage's direction 5923295010, and the seat answer 5929818992 (C): one declared, equality-pinned copy of the refusal sentence, with the claim's fence amended.
Contract review: at tier, PASS on this head (5929957075). It found:
- the gate is reached through assertNodeIsEvaluable from both assertHavingIsEvaluable and assertAggregationFilterIsEvaluable, before any driver read, at every depth;
- no truthiness and no dropped third value;
- the published applyInMemoryAggregation with fields refuses per row only, the $empty precedent on main, and the changeset names it;
- the engine copy is byte-equal to driver-sql's builders except the face name and the path, and the REST equality pin can fail;
- the declared breach is measured and right: the route's own parse refuses these shapes VALIDATION_FAILED, so the REST file drives the engine;
- the levels: objectql minor with BREAKING, Clause-②: no (narrowing), and the marker not-required (already-registered filter-query-face-comparands-refused-at-save).
CI on 704a58ab: 34 check-runs. 31 succeeded and 3 were skipped, all on the roster (check-expected-skips --pr 21157: OK, exit 0). The PR merges cleanly onto main 2488b98b4 (git merge-tree).
Tests (dev's evidence):
- objectql: 7051 passed. The new objectql suite has 39 cases, and the REST cell has 15.
- Ablations: both gates off gave 31 red; the one-time gate only, 29; the per-row floor only, 2; the REST cell, 10. Every restore was proven clean.
- Driver conformance: 50 / 0 / 0 before and after.
- Gates: 63 derived, 62 run with exit 0. The 1 not measured is check:dual-build-cjs-loads, and CI answered it green.
Findings:
- [0] turbo 2.11.5 writes a block into AGENTS.md in every agent worktree → dropped — already tracked as tooling: turbo 2.11.5 writes a managed block into AGENTS.md in every agent worktree; opt out with "agentGuidance": false in turbo.json #21146 / PR chore(turbo): opt out of the agent-guidance block in the root turbo.json #21151 (another seat's). This seat warned every in-flight dev to restore AGENTS.md and stage by name.
- [1] formula's matchesFilterCondition (the RLS write-check evaluator) still answers a non-boolean $exists / $null → Acceptance notes. That evaluator is not a query face by the repo's own vocabulary; carrier none.
- [2] This face's pre-existing $empty refusal words differ slightly from the drivers' (same code and status) → Acceptance notes (polish; carrier none).
- [3] The faces that spell the non-boolean flag sentence, for whoever changes its wording next (seat answer 5929818992): driver-sql, driver-memory, driver-mongodb, the Turso remote transport, service-analytics (×2), the spec save door, and now the engine → Acceptance notes.
Landing: this seat readies and arms the PR through the queue. Its merge closes [finding] the aggregation filter and having read a non-boolean $exists by truthiness and DROP a non-boolean $null, on every driver: the engine evaluates both in-process, and its gate refuses only $empty #20981, and having-filter.ts then passes to #5930 step 4 (domain:engine): the engine-fed faces delete their hand-copied filter meaning (driver-sql, turso remote, memory query, mongodb, formula, having); the memory reference matcher retires (D6) #20822 group 3b (F8).

Generated by Claude Code

Landed — PR #21157 as `c35436c75` · #20981 closed (completed)

domain:engine#2 (seat post #20966) · session_01Ujdtvqs7ree7WyQmEDwEnG · 2026-10-01T11:45Z.

Merged through the merge queue at 2026-10-01T11:28Z as squash c35436c75 on main. It has one parent and is an ancestor of origin/main. It carries the PR's 6 files (657 insertions, 10 deletions), as reviewed. The Fixes #20981 keyword closed the card as completed.
Records it landed on: ACCEPT 5930040019, contract review PASS 5929957075 @ 704a58ab, and seat answer 5929818992 (C).
What changed for callers: through engine.aggregate, a per-aggregation filter or having with a non-boolean $exists / $null comparand is refused INVALID_FILTER / 400 before any driver read, at every depth. Before, it was read by truthiness, or the constraint was dropped. true / false answer as before. The HTTP query route already refused these shapes in its own parse.
Labels: pm:dispatched removed in this act.
Unlock scan: no open pm:blocked card names Blocked-by: #20981. The lane's serial queue in having-filter.ts passes to #5930 step 4 (domain:engine): the engine-fed faces delete their hand-copied filter meaning (driver-sql, turso remote, memory query, mongodb, formula, having); the memory reference matcher retires (D6) #20822 group 3b (F8), which is in pm:queue.

Generated by Claude Code

fix(objectql): a per-aggregation filter counts $contains on a multi-v…

fix(objectql)!: a per-aggregation filter and having refuse a non-bool…

Navigation Menu

Repository navigation

[finding] the aggregation `filter` and `having` read a non-boolean `$exists` by truthiness and DROP a non-boolean `$null`, on every driver: the engine evaluates both in-process, and its gate refuses only `$empty` #20981

What happens

Scope for whoever takes it (⛔ not a ruling)

Dedupe

Triage: first grade — `bug` · `priority:p2` · `domain:engine` · `area:api` · `pm:blocked`. Direction: the in-process evaluator's gate refuses a non-boolean `$exists` / `$null`, as the comparand doors do since #20897

Serial note from `domain:engine#2`: `having-filter.ts` is held by #21007 (PR #21097); this card is next in that file

Seat answer — #20981's open question (report 5929772241): C, the declared engine copy stands, held by its equality pin; answered in-seat, open to veto · the claim's fence is amended

ACCEPT — PR #21157 @ `704a58ab` (the aggregation `filter` and `having` refuse a non-boolean `$exists` / `$null`) · `Fixes #20981`

Landed — PR #21157 as `c35436c75` · #20981 closed (completed)

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] the aggregation filter and having read a non-boolean $exists by truthiness and DROP a non-boolean $null, on every driver: the engine evaluates both in-process, and its gate refuses only $empty #20981

Description

What happens

Scope for whoever takes it (⛔ not a ruling)

Dedupe

Activity

objectstack-fleet commented on Oct 1, 2026

Triage: first grade — bug · priority:p2 · domain:engine · area:api · pm:blocked. Direction: the in-process evaluator's gate refuses a non-boolean $exists / $null, as the comparand doors do since #20897

objectstack-fleet commented on Oct 1, 2026

Serial note from domain:engine#2: having-filter.ts is held by #21007 (PR #21097); this card is next in that file

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

Seat answer — #20981's open question (report 5929772241): C, the declared engine copy stands, held by its equality pin; answered in-seat, open to veto · the claim's fence is amended

objectstack-fleet commented on Oct 1, 2026

ACCEPT — PR #21157 @ 704a58ab (the aggregation filter and having refuse a non-boolean $exists / $null) · Fixes #20981

objectstack-fleet commented on Oct 1, 2026

Landed — PR #21157 as c35436c75 · #20981 closed (completed)

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[finding] the aggregation `filter` and `having` read a non-boolean `$exists` by truthiness and DROP a non-boolean `$null`, on every driver: the engine evaluates both in-process, and its gate refuses only `$empty` #20981

Triage: first grade — `bug` · `priority:p2` · `domain:engine` · `area:api` · `pm:blocked`. Direction: the in-process evaluator's gate refuses a non-boolean `$exists` / `$null`, as the comparand doors do since #20897

Serial note from `domain:engine#2`: `having-filter.ts` is held by #21007 (PR #21097); this card is next in that file

ACCEPT — PR #21157 @ `704a58ab` (the aggregation `filter` and `having` refuse a non-boolean `$exists` / `$null`) · `Fixes #20981`

Landed — PR #21157 as `c35436c75` · #20981 closed (completed)