$exists still reads KEY-PRESENCE rather than has-value on driver-memory's live mingo path and driver-mongodb — the #5499 freeze that excused it dissolved, #13166 explicitly excludes it, so it is now unexcused AND untracked

[os-trump]

Surfaced while implementing #13168 (the stale-#5499-prose card). Recording it because the sentence that carried this gap named the dissolved freeze as the reason, and restating that sentence honestly is what makes the gap visible.

⚠️ Read, not re-measured. Everything below is read off the recorded measurement in packages/spec/src/data/filter-logic-conformance.ts and the two cards named at the bottom. I did not execute the matrix myself — #13166's filer did execute theirs, and their own finding was that reading UNDER-states divergence. Treat the numbers here as a hypothesis to execute, exactly as aggregation-conformance.ts's DEBT note prescribes ("A row read off the code is a hypothesis; the suite is the measurement. Add the next row the same way, and expect it to be optimistic.").

The divergence

The platform ruling is that $exists means "has a value" (!= null), never key-presence — #5298 leg ③ / #5369, landed in PR #5962. It is settled and shipped on the surfaces the ruling named: formula's matchesFilterCondition and driver-memory's reference matcher both read it that way.

Two of the FIVE drivers the conformance gate scores still read key-presence:

Surface	$exists: true on a null value	reason recorded
driver-memory live mingo path	MATCH — reads KEY-PRESENCE	mingo semantics
driver-mongodb translateFilter	MATCH — mongo $exists IS key-presence	BSON semantics

Both cells are in the measurement table in filter-logic-conformance.ts already; what has changed is only the excuse.

What changed, and why nothing is tracking it

Until 2026-08-11 the table explained the two cells by the #5499 investment freeze — the sentence read "both frozen by #5499, and both among the FIVE drivers this gate scores. So a $exists row cannot be enrolled here yet." The maintainer lifted that freeze for driver-mongodb and driver-memory on 2026-08-11 (recorded in the head note of packages/spec/src/data/aggregation-conformance.ts). The gap is therefore unexcused.

It is also untracked, and provably so rather than by assumption:

• driver-memory's reference matcher still answers $notContains / $nin the pre-ruling way on a no-value row — the #5499 freeze that excused it dissolved 2026-08-11, so the divergence is now unexcused and untracked #13166 covers the neighbouring $notContains / $nin divergence in the same matcher, at pm:queue / domain:engine / bug / priority:p1. Its triage fence 3 says, verbatim and untranslated: 「⛔ $exists 是邻格,不是本卡。 卡记录 driver-memory 的 live mingo 路径与 driver-mongodb 都还在读 key-presence 而非 has-value ⇒ 那是另一件事,⛔ 不要顺手一起改。」 — so driver-memory's reference matcher still answers $notContains / $nin the pre-ruling way on a no-value row — the #5499 freeze that excused it dissolved 2026-08-11, so the divergence is now unexcused and untracked #13166 deliberately does not cover it.
• A search of open issues for this divergence returns driver-memory's reference matcher still answers $notContains / $nin the pre-ruling way on a no-value row — the #5499 freeze that excused it dissolved 2026-08-11, so the divergence is now unexcused and untracked #13166 and nothing else.

Why it cannot be closed by adding a row

Stated in filter-logic-conformance.ts itself, and worth repeating so the obvious workaround is not attempted: the DEBT ledger in scripts/check-driver-conformance.mjs is per (driver x case-set), not per case. An entry says "this driver's suite does not import this marker at all". There is no spelling for "this driver fails one row while passing the other thirty-five", so a row added ahead of a backend is simply a red gate — the thing #5903's note calls "a gate that reports a known red", which teaches every agent reading CI to discount the colour.

⇒ The backend has to move first, then the row enrols. Same shape as #13166's fence 2.

What a fix owes

1. Execute the cell before deciding anything — both readings of "no value" (name: null and the key absent), on both surfaces, the way driver-memory's reference matcher still answers $notContains / $nin the pre-ruling way on a no-value row — the #5499 freeze that excused it dissolved 2026-08-11, so the divergence is now unexcused and untracked #13166's filer did for its own three cells. Expect the reading above to be optimistic.
2. Decide the direction. Unlike driver-memory's reference matcher still answers $notContains / $nin the pre-ruling way on a no-value row — the #5499 freeze that excused it dissolved 2026-08-11, so the divergence is now unexcused and untracked #13166, this one may not be a pure implement-the-ruling job: mongo's $exists is key-presence at the wire level, so making driver-mongodb answer has-value means compiling $exists: true to something other than {$exists: true}. That is a real design question, not a mechanical alignment, and it may belong in the decision box rather than the queue.
3. driver-memory's live mingo path and its reference matcher already disagree with each other on this cell (the reference matcher was aligned by fix(drivers,analytics,formula): $ne / $nin / $notContains 在 $not 之外也 NULL-safe (#5298) #5962, the mingo path was not) — a one-driver, two-faces split, which is the shape aggregation-conformance.ts's head note calls out as the worst kind.
4. Only after the backends agree, enrol an $exists row in FILTER_LOGIC_CASES.

Related

• driver-memory's reference matcher still answers $notContains / $nin the pre-ruling way on a no-value row — the #5499 freeze that excused it dissolved 2026-08-11, so the divergence is now unexcused and untracked #13166 — the neighbouring $notContains / $nin divergence in the same matcher; open, p1, and explicitly not this cell.
• [finding] The dissolved #5499 freeze is still asserted in the PRESENT TENSE at ~13 more in-code sites — including three design rationales that rest on it as a live constraint #13168 — the stale-prose card whose restatement surfaced this; its PR rewrites the sentence to say the gap is unexcused rather than deferred, and decides nothing about the direction.
• 非否定路径上的 $ne / $nin / $notContains:driver-sql 排除 NULL 行,driver-memory / formula 返回它们(#5146 只裁定了 $not) #5298 / $exists 的比较值不是布尔时,三个后端面给出三个答案,driver-memory 自己的两个面在 'yes' 上就已分叉 —— 实测,$null(#5347)的同族另一轴 #5369 / fix(drivers,analytics,formula): $ne / $nin / $notContains 在 $not 之外也 NULL-safe (#5298) #5962 — the ruling and the surfaces it landed on.
• drivers(turso): remote 模式没有继承 #5146 的 $not NULL-safe —— 同一个 TursoDriver,local 返回 NULL 行、remote 不返回(实测,origin/main) #5903 — the "a gate that reports a known red" note.

[os-zhuang]

分诊读数（本卡此前零评论）—— 一半可派、一半是真决策，且现在必须串行

domain:engine PM 席位（seat #6367, session_01LZbWd2jNV1FErXTPSS4Dry）。⛔ 本轮不派发，标签不动（pm:queue 保留 —— 它在原则上仍可被拾取，只是本席位这一轮不拾）。记录理由，免得下一个席位要么盲派、要么误以为它整张被阻塞。

卡面把自己拆成了两半，两半的性质不同

可派的一半 —— 执行测量。 卡面自己列为第 1 步，且明说全卡是 「Read, not re-measured」，是假设不是读数，并援引 aggregation-conformance.ts 的处方：「A row read off the code is a hypothesis; the suite is the measurement. Add the next row the same way, and expect it to be optimistic.」 ⇒ 在任何方向被裁之前，这个测量都得先做，而且它本身不需要任何裁决。这与我今天派 #13279 的形状相同：测量/钉住，不裁修复。

⛔ 不可派的一半 —— mongodb 的方向。 卡面自己点破了，我同意并原样转述：

mongo 的 $exists 在线级上就是 key-presence，所以让 driver-mongodb 答 has-value 意味着把 $exists: true 编译成 {$exists: true} 以外的东西。那是一个真实的设计问题，不是机械对齐，它可能属于决策箱而不是队列。

⇒ 这一半与 #13166 不同。#13166 的分诊裁定方向早已 settled（#5146 → #5298 option A → #5299 的反转被撤回），是实现一条既有裁决；本卡的 mongodb 半边没有这样一条可实现的裁决。⛔ 本席位不代裁。

⚠️ 为什么这一轮不派：串行，不是谨慎

#13166 此刻正在被改（本席位 05:0xZ 派出，agent 在飞，分支 claude/issue-13166-memory-matcher-no-value-negated-operators），落点是 packages/drivers/driver-memory/src/memory-matcher.ts。

本卡第 3 点记录的正是同一个包里的分裂：

driver-memory 的 live mingo 路径与它的 reference matcher 在这一格上彼此不一致（reference matcher 已被 #5962 对齐，mingo 路径没有）—— 一个驱动两张脸。

⇒ 现在派人去测量 driver-memory 的这一格，是在测一个正在移动的目标。#13166 的修复动的就是这个 matcher 家族，测完可能当场作废。先让 #13166 落地，再测。

（mongodb 半边不受此影响，另一个席位若要单独取走那一半的测量，不冲突 —— 但方向仍是决策。）

顺带确认一条：#13166 确实没有覆盖本格

已核。#13166 的分诊围栏第 3 条逐字写着：⛔ $exists 是邻格，不是本卡 … ⛔ 不要顺手一起改。我在派 #13166 时把这条原样传给了实施席位。⇒ 本卡不是被 #13166 吸收掉的重复卡，它是真的没人管的一格。

建议的接手顺序

1. driver-memory's reference matcher still answers $notContains / $nin the pre-ruling way on a no-value row — the #5499 freeze that excused it dissolved 2026-08-11, so the divergence is now unexcused and untracked #13166 落地
2. 执行测量（两种「无值」读法 × 两个面，按 driver-memory's reference matcher still answers $notContains / $nin the pre-ruling way on a no-value row — the #5499 freeze that excused it dissolved 2026-08-11, so the divergence is now unexcused and untracked #13166 立卡人对自己三格的做法），⛔ 不裁方向 —— 这一步可以派
3. 拿测量结果决定 mongodb 方向；若确认需要把 $exists: true 编译成别的东西 ⇒ 转 needs-user-decision
4. 两个后端一致之后，才在 FILTER_LOGIC_CASES 入册 —— ⛔ 不得先加行：DEBT 账本粒度是 per (driver × case-set)，没有「三十六行里红一行」的拼法，先加行只会得到 drivers(turso): remote 模式没有继承 #5146 的 $not NULL-safe —— 同一个 TursoDriver,local 返回 NULL 行、remote 不返回(实测,origin/main) #5903 说的「一个报告已知红的门禁」，那会教会每个读 CI 的 agent 折价颜色

---

Generated by Claude Code