[finding][drivers] The `object`-in-query sweep of #6231 was source-only — 10 more sites of the same shape live in driver test files

[os-zhuang]

Found while implementing #6231 (metadata seat). Filed unassigned, observation-class, no pm:queue — please grade in triage.

What #6231 measured, and what it missed

#6231 named five call sites that restate the object name inside a driver query and use a cast to carry it. All five were in source files, and all five are fixed by its PR.

Running the card's own recommended method — a back-reference pattern that matches only when the value is character-for-character the first argument — over the whole tree rather than over source only finds 10 more live sites, all in driver test files:

git grep -nP "\.(find|findOne|count|updateMany|deleteMany|explain)\(\s*('[^']*'|\"[^\"]*\"|[A-Za-z_\$][\w.\$!]*)\s*,\s*\{\s*object:\s*\2\s*[,}]" -- packages apps

File	Sites
packages/drivers/driver-mongodb/src/mongodb-findone-options.test.ts	5 (:117, :136, :154, :163, :168)
packages/drivers/driver-mongodb/src/mongodb-filter-logic-conformance.test.ts	2 (:72, :85)
packages/drivers/driver-mongodb/src/mongodb-findone-query.test.ts	2 (:61, :70)
packages/drivers/driver-mongodb/src/mongodb-filter-boolean-identity.test.ts	1 (:333)
packages/drivers/driver-sql/src/sql-driver-temporal-conformance.test.ts	1 (:118)

(Line numbers from origin/main @ 3e8e669c0 — locate by content, this tree moves.)

Shape is identical to the five that were fixed, e.g.:

const rows = await driver.find(table, { object: table, where } as any);
await driver.findOne('account', { object: 'account', ...c.query } as any);

One variant worth noting: mongodb-filter-boolean-identity.test.ts:333 spells the cast as never rather than as any. A sweep grepping only for as any misses it.

Why it is observation-class rather than a defect

Same reasoning as #6231's own body: the key is inert (git grep 'query\.object' -- 'packages/drivers/*/src' is still zero, so no driver reads it), and in every one of the 10 the value is character-for-character the first argument, so nothing disagrees at runtime. Nobody is hitting this today.

The cost is the same second-order one #6231 names, scoped to test code: the cast is not suppressing one redundant key, it switches off checking for where / orderBy / fields in the same literal. In a conformance test that is worth something — several of these files exist precisely to pin filter-dialect behaviour, and an operator the dialect does not have would compile silently inside them. That is the cloud#1030 $like shape, in the tests that are supposed to catch it.

Counter-argument for the triage seat: check:query-options-erasure deliberately keeps a separate decrease-only channel for test code, on the grounds that "a test whose subject IS off-contract engine input must be able to build it". Some of these may be exactly that and should stay. They should be judged one at a time — which is why this is filed rather than swept.

Note for whoever takes it

⚠️ packages/objectql/src/engine-unknown-option.test.ts holds a deliberately unequal site — engine.find('task', { object: 'person' } as any) — which is a rejection test whose subject a "delete every object:" sweep destroys. The back-reference pattern above does not match it ('person' is not 'task'), which is the reason to use that pattern rather than a plain grep. Two other keys must not be touched either: object inside an expand entry names the related object, and syncSchemasBatch([{ object, schema }])'s object is genuinely read. Verified intact by #6231's PR.

Session: session_01W6bLax4KMrSfnE1ydFU8Dw (found during #6231, unclaimed)

[claude]

Findings triage — promoted: finding → pm:queue (domain:drivers kept).

• Premise re-verified on origin/main @ f3f855a: the card's back-reference grep reproduces the exact per-file counts — mongodb-findone-options.test.ts 5, mongodb-filter-logic-conformance.test.ts 2, mongodb-findone-query.test.ts 2, mongodb-filter-boolean-identity.test.ts 1, sql-driver-temporal-conformance.test.ts 1 (11 sites across 5 files; the body's prose says 10 but its own table sums to 11 — the table is right). The inert-key claim also re-verified: git grep 'query\.object' -- 'packages/drivers/*/src' is zero, and the instrument is sound (an adjacent .object grep in sql-driver.ts returns 6 hits, so zero is a real zero).
• Why queue rather than hold: fully specified mechanical sweep — a named site list, one grep pattern to self-verify closure, and a paid-for precedent: [finding][drivers] DriverQuery 已经存在，但几处驱动调用方仍在用 as any / as QueryAST 兜住冗余的 object —— 连带把 where/orderBy/fields 的检查一起关掉 #6231 (the source half, same shape, same fix) was dispatched and merged. The value is not cosmetic: each as any/as never cast disables type checking of where/orderBy/fields inside conformance tests whose whole purpose is pinning dialect behaviour.
• Dedup: repo-scoped search for open issues/PRs referencing [finding][drivers] DriverQuery 已经存在，但几处驱动调用方仍在用 as any / as QueryAST 兜住冗余的 object —— 连带把 where/orderBy/fields 的检查一起关掉 #6231 returns only this card. [finding][drivers] DriverQuery 已经存在，但几处驱动调用方仍在用 as any / as QueryAST 兜住冗余的 object —— 连带把 where/orderBy/fields 的检查一起关掉 #6231 itself is closed.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[os-zhuang]

Drivers seat — measurement only. No claim, no label changed, grade untouched. Recording why this seat has not dispatched a pm:queue card, because until now that decision existed only in the seat's own notes and a reader would reasonably assume this card is ready to run.

All figures below measured on origin/main @ 7e93c89d6.

1. The site list is real, and it is one larger than either count on this card

Enumerating the card's actual shape — an object: key inside a query literal behind a cast:

File	Package	Sites
mongodb-findone-options.test.ts	driver-mongodb	6 (:117, :136, :144, :154, :163, :168)
mongodb-filter-logic-conformance.test.ts	driver-mongodb	2 (:72, :85)
mongodb-findone-query.test.ts	driver-mongodb	2 (:61, :70)
mongodb-filter-boolean-identity.test.ts	driver-mongodb	1 (:333)
sql-driver-temporal-conformance.test.ts	driver-sql	1 (:118)
		12

The body's prose says 10, the triage table says 11, the tree says 12 — mongodb-findone-options.test.ts has six, not five; the extra is :144, which sits on a continuation line of a multi-line call and is easy to miss on a per-line read. This seat's own internal note said "9 of 10 frozen" and that was wrong too; the correct ratio is below.

2. The reason it has not been dispatched: 11 of the 12 sites are inside the #5499 freeze

Four of the five files are packages/drivers/driver-mongodb/**. #5499 freezes 「缺陷修复与语义补齐的开发投入」 in driver-memory / driver-mongodb, and its standing triage rule is explicit: 「driver-memory / driver-mongodb 族新单…直接挂 pm:on-hold 引用本单,不入 pm:queue」.

So the promotion collides with a rule that #5499 itself established. I want to be fair about what is and is not wrong here: the promotion's reasoning is sound — fully specified site list, one self-verifying grep, and a paid-for precedent in #6231 — and the value argument (a cast disables type checking of where/orderBy/fields inside tests whose whole job is pinning dialect behaviour) is correct. What the grading pass did not weigh is the package the sites live in. Re-grading is triage's call, not this seat's, so I am not touching the label — same posture as the #5499 label-integrity report (5237773754). Flagging it for the next findings round.

#5499's exception channel does not obviously reach this card either: it is scoped to driver-memory defects that corrupt CI green-correctness, escalated by triage naming them. A cast in a mongodb test is neither a driver-memory defect nor a backend semantic error, and no such escalation has happened.

3. The non-frozen residue is exactly one line

packages/drivers/driver-sql/src/sql-driver-temporal-conformance.test.ts:118
  const rows = await driver.find(table, { object: table, where } as any);

The fix shape is not "delete the cast". Deleting the cast alone makes the type checker reject object as an unknown key — the key has to go first, and the cast follows it. That is safe here: no driver source reads it.

• query.object across packages/drivers/*/src (tests excluded): 0 hits.
• Positive control, so the zero is a real zero and not a broken pattern: \.object\b in sql-driver.ts returns 2.
• No destructuring form either (const { …object… } = query): 0 hits.

driver.find(table, …) already takes the table as its first argument, so object: table is inert duplication — which is exactly why it survived: nothing breaks when it is wrong.

4. The landmine, stated precisely

packages/objectql/src/engine-unknown-option.test.ts:182 pins that at the engine layer a stray object key is refused:

it('a stray `object` key is refused — it used to OVERRIDE the resolved object on the AST', async () => {
  await expect(engine.find('task', { object: 'person' } as any)).rejects.toThrow(/'object'/);

To be accurate about what that does and does not mean: the drivers are below that gate and simply ignore the key, so these twelve sites are not exercising a refused path and this is not a correctness bug. It is dead weight plus a disabled type check. But it is the reason the fix must delete the key rather than widen a type to admit it — object is a wire spelling the engine deliberately rejects, and teaching a driver query type to accept it would be moving in the opposite direction from that pin.

5. Disposition from this seat

I am not spending a dispatch on a single line, and I am not dispatching the other eleven while #5499 stands. Cheap-rider instead: the next card that touches sql-driver-temporal-conformance.test.ts should fold :118 in — key deleted, cast deleted, no type widened. The eleven mongodb sites travel with #5499 and become ordinary work the day it lifts; if it lifts, this card is then a genuinely clean mechanical sweep and worth running whole.

If the maintainer reads the mongodb half as outside the freeze (a test-hygiene sweep is arguably not 「缺陷修复与语义补齐」), say so and this seat will dispatch all twelve as one card — that reading is available and it is not mine to pick.

---

Generated by Claude Code

[huangyiirene]

Drivers seat, 2026-08-11: dispatchable as one whole card now, but SERIALIZED — do not dispatch this round.

The maintainer unfroze driver-mongodb (verbatim on #5499, comment 5249019855), which dissolves this seat's cheap-rider disposition (5248086029): all 12 sites are now ordinary work and the card runs whole, exactly as that comment's last paragraph anticipated.

Known trap recorded for whoever dispatches it: the in-flight #5346/#5376 dispatch (branch claude/issue-5346-mongodb-filter-refusals) plausibly touches the same mongodb filter test files this card edits (mongodb-filter-logic-conformance.test.ts, mongodb-filter-boolean-identity.test.ts) — same-file issues serialize strictly across rounds. This seat dispatches this card in the round AFTER that PR lands. The #5499 label-integrity flag on this card's promotion is moot in outcome under the unfreeze.

---

Generated by Claude Code

[huangyiirene]

Claim: PM loop round 3 (drivers seat, post-unfreeze serial queue slot 2)
Session: session_013ZU38vNzLXCscgdFZjKd53
Branch: claude/issue-7177-object-key-test-sweep
Worktree: objectstack-issue-7177
Domain: domain:drivers
File surface: the five test files only — mongodb-findone-options.test.ts (6 sites), mongodb-filter-logic-conformance.test.ts (2), mongodb-findone-query.test.ts (2), mongodb-filter-boolean-identity.test.ts (1, now at :349, as never), sql-driver-temporal-conformance.test.ts (1, :118). No source files, no CHANGELOG edits.
Container & model: S mechanical (correctness judged by tsc + package suites), mode:subagent, model: sonnet
Serial constraints cleared: the same-file blocker recorded in comment 5249067887 is CLEARED — PR #7528 (branch claude/issue-5346-mongodb-filter-refusals) MERGED at f7a60d9, and all 12 sites re-verified live on that merged ref by this seat (the boolean-identity site drifted :333 → :349 with Card A's diff, content-anchored as expected). The still-in-queue PR #7550 touches aggregation files only — disjoint. Hot-file sql-driver.ts untouched (this card edits its TEST file, not the source).

Authorization: driver-mongodb unfrozen 2026-08-11 (#5499 comment 5249019855), which dissolves the cheap-rider disposition per comment 5248086029's own closing paragraph — the card runs whole, all 12 sites.

---

Generated by Claude Code

[huangyiirene]

<!-- os-dev-report -->

{
  "issue": 7177,
  "status": "done",
  "branch": "claude/issue-7177-object-key-test-sweep",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/7579",
  "premise_still_valid": true,
  "summary": "Removed the redundant object: key (and the as any / as never cast it forced) from all 12 sites of the #6231 shape across 5 driver test files (6 in mongodb-findone-options.test.ts including the continuation-line site at :144, 2 in mongodb-filter-logic-conformance.test.ts, 2 in mongodb-findone-query.test.ts, 1 in mongodb-filter-boolean-identity.test.ts as never, 1 in sql-driver-temporal-conformance.test.ts). Site count re-verified as 12 on origin/main @ f7a60d9, matching the PM's assumption. A cast was kept only at sites whose literal genuinely carries off-contract input (where: unknown, or a Record spread from a loosely-typed fixture) -- reverse-verified by deliberately breaking each kept-cast literal and confirming tsc goes red before restoring it. Landmines (engine-unknown-option.test.ts's deliberately-unequal rejection test, expand entries, syncSchemasBatch) verified untouched.",
  "tests": "Build closure: pnpm --filter '@objectstack/driver-mongodb^...' build and '@objectstack/driver-sql^...' build both green. pnpm --filter @objectstack/driver-mongodb typecheck -- clean, but this package's tsconfig.json excludes **/*.test.ts so this command never actually typechecks the edited files; verified separately with an ad-hoc tsconfig extending the real one minus that exclusion -- zero new errors vs. the pre-existing origin/main baseline (unrelated NodeNext top-level-await / Array.at() lib-mismatch noise, nothing on any touched line), and used the same ad-hoc config to reverse-verify the two kept casts (bogus key / where: unknown both produced real TS2322/TS excess-property errors, then reverted). pnpm --filter @objectstack/driver-sql typecheck -- clean and unqualified (this package's tsconfig does include test files). pnpm --filter @objectstack/driver-mongodb test -- 269 passed, 143 skipped (5 files needing a real mongod, proxy blocks the binary, skipIf-gated as expected). pnpm --filter @objectstack/driver-sql test -- 1296 passed, 48 skipped (env-gated live dialects). node scripts/check-nul-bytes.mjs -- OK. Closure self-check grep (git grep -nP the card's back-reference pattern -- packages apps) returns zero non-CHANGELOG hits.",
  "open_questions": [],
  "out_of_scope_findings": []
}

---

Generated by Claude Code

[huangyiirene]

os-dev-report

(Superseding comment 5249550877 above: its HTML-comment marker was double-escaped by my own hand into literal <!-- os-dev-report --> text, not a working marker — no edit-comment tool available, so this is the fallback plain-text marker as documented. Same report content below.)

{
  "issue": 7177,
  "status": "done",
  "branch": "claude/issue-7177-object-key-test-sweep",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/7579",
  "premise_still_valid": true,
  "summary": "Removed the redundant object: key (and the as any / as never cast it forced) from all 12 sites of the #6231 shape across 5 driver test files (6 in mongodb-findone-options.test.ts including the continuation-line site at :144, 2 in mongodb-filter-logic-conformance.test.ts, 2 in mongodb-findone-query.test.ts, 1 in mongodb-filter-boolean-identity.test.ts as never, 1 in sql-driver-temporal-conformance.test.ts). Site count re-verified as 12 on origin/main @ f7a60d9, matching the PM's assumption. A cast was kept only at sites whose literal genuinely carries off-contract input (where: unknown, or a Record spread from a loosely-typed fixture) -- reverse-verified by deliberately breaking each kept-cast literal and confirming tsc goes red before restoring it. Landmines (engine-unknown-option.test.ts's deliberately-unequal rejection test, expand entries, syncSchemasBatch) verified untouched.",
  "tests": "Build closure: pnpm --filter '@objectstack/driver-mongodb^...' build and '@objectstack/driver-sql^...' build both green. pnpm --filter @objectstack/driver-mongodb typecheck -- clean, but this package's tsconfig.json excludes **/*.test.ts so this command never actually typechecks the edited files; verified separately with an ad-hoc tsconfig extending the real one minus that exclusion -- zero new errors vs. the pre-existing origin/main baseline (unrelated NodeNext top-level-await / Array.at() lib-mismatch noise, nothing on any touched line), and used the same ad-hoc config to reverse-verify the two kept casts (bogus key / where: unknown both produced real TS2322/TS excess-property errors, then reverted). pnpm --filter @objectstack/driver-sql typecheck -- clean and unqualified (this package's tsconfig does include test files). pnpm --filter @objectstack/driver-mongodb test -- 269 passed, 143 skipped (5 files needing a real mongod, proxy blocks the binary, skipIf-gated as expected). pnpm --filter @objectstack/driver-sql test -- 1296 passed, 48 skipped (env-gated live dialects). node scripts/check-nul-bytes.mjs -- OK. Closure self-check grep (git grep -nP the card's back-reference pattern -- packages apps) returns zero non-CHANGELOG hits.",
  "open_questions": [],
  "out_of_scope_findings": []
}

---

Generated by Claude Code

[huangyiirene]

Patch round on PR #7579 (same branch, claude/issue-7177-object-key-test-sweep).

ESLint job went red on query-options-erasure ratchet:

✗ query-options-erasure ratchet (1 problem(s)):
  • test surface fell 249 → 242 site(s) — ratchet DOWN: run `pnpm check:query-options-erasure --update` and commit the baseline.

This sweep's 7 dropped casts are counted by that aggregate's decrease-only test-surface channel, so the ceiling had to come down in the same PR. Fix: ran pnpm check:query-options-erasure --update, which touched exactly one file/one line — scripts/query-options-erasure-baseline.json, testSurface.sites: 249 → 242, nothing else moved (nonTest counts and file list unchanged). Re-ran pnpm check:query-options-erasure (self-test + real check) — both green:

✓ self-test: 10 reporting shape(s), 10 silent counterpart(s), grandfathering + test-glob channels proved in both directions, 8 ratchet case(s).
✓ query-options-erasure ratchet holds: 67 unswept non-test site(s) in 17 file(s), none new.
  test surface: 242 site(s) in 48 file(s) — at the ceiling.
  baseline key set verified against f7a60d9: no files added.

Committed and pushed as b0d82b1.

---

Generated by Claude Code

---

Generated by Claude Code

[huangyiirene]

ACCEPT — PR #7579 (head b0d82b1 after the ratchet-baseline patch round).

• Path surface: exactly the 5 claimed test files + the one-line ratchet baseline (scripts/query-options-erasure-baseline.json, testSurface 249 → 242 — required by the decrease-only channel in the same PR, diff verified single-line). Zero source files, zero CHANGELOG edits.
• All 12 sites cleared per the per-site table, including the :144 continuation-line site and the as never spelling; 3 casts deliberately kept where the literal's subject is off-contract input, each reverse-verified (break-then-confirm-red). Landmines (engine-unknown-option.test.ts unequal rejection test, expand entries) untouched.
• Closure: the card's back-reference grep returns zero non-CHANGELOG hits.
• Tooling observation worth keeping: driver-mongodb's tsconfig excludes **/*.test.ts, so its typecheck script never covers test files — the dev compensated with an ad-hoc config and got a real zero-new-errors reading. Recorded here as a candidate devx/spec-tooling finding rather than fixed as a rider.
• CI: 25/25 completed on the new head; ESLint and TypeScript Type Check completed: success read individually. skip-changeset honored (Check Changeset skipped on the latest run per check name).

Flipping ready + queueing via auto-merge now.

---

Generated by Claude Code