$exists docs and two PUBLISHED skills still teach key-presence — false since the has-value alignment landed, and no gate can catch it because the teaching corpus does not execute

Blocked-by: #13709
Unlock-action: re-check the JSDoc at packages/spec/src/data/filter.zod.ts:954

Six places in the teaching corpus still describe $exists as key-presence. That stopped being true when the has-value alignment landed at 9dac1ae017 (PR #13529), so these lines are not stale wording — they are false, and two of them ship to customers.

Found by the consumer census on #13492 (comment 5470462097). Measured on origin/main @ 430a278371.

The six false sites

site	what it says today	why it is false
`packages/spec/src/data/filter.zod.ts:954`	JSDoc: "Field exists check (primarily for NoSQL) - MongoDB: `$exists`"	⚠️ governed surface — see the fence note below
`content/docs/protocol/objectql/query-syntax.mdx:264`	table: "Field exists (NoSQL)"	operator means has-value on every backend
`content/docs/protocol/objectql/query-syntax.mdx:518`	code block commented `// Field exists (NoSQL)`	same, and it carries an `os:check` marker
`content/docs/protocol/kernel/http-protocol.mdx:397`	"Field existence check"	same
`skills/objectstack-query/SKILL.md:157`	"Field exists (NoSQL) / MongoDB `$exists`"	published; `translateFilter` no longer emits MongoDB `$exists`
`skills/objectstack-query/rules/filters.md:23`	"Existence / (NoSQL) `$exists`"	published

⛔ One site is already CORRECT — do not sweep it

content/docs/data-modeling/queries.mdx:71 reads "Field has a value", which is right and matches the shipped behaviour. Naming only the wrong sites invites a blanket find-and-replace that would break the one line already telling the truth. Whoever takes this should treat that row as the target wording, not as another thing to change.

The sharpest one, and why this is urgent rather than tidy

skills/objectstack-query/SKILL.md:157 tells authors that $exists maps to MongoDB $exists. Since 9dac1ae017, driver-mongodb's translateFilter lowers it to {$ne: null} / {$eq: null} and never emits MongoDB's $exists at all. A customer following that line today writes a filter whose behaviour does not match what the skill promises — on a null-valued field the two readings return different rows. The skills/ directory is the published, customer-facing surface, and this should be graded on that basis rather than as an internal docs nit.

⭐ Why no gate caught this, and why none would have

This is the reusable part, and it is a statement about the gate's reach rather than about whoever landed the alignment.

PR #13529 touched 11 files: driver code and driver tests only — no docs, no spec, no skills, no examples. Its CI was green and correctly so: every executable assertion about $exists was updated in the same commit. The teaching corpus does not execute. No test imports a .mdx table row, and no gate reads a skill's prose for semantic agreement with the drivers it describes, so there is no arrangement of the current gates under which this would have gone red.

That means the drift is silent by construction and will stay wrong until someone reads it — which is how this class of defect survives for months. Worth considering, separately from the repair, whether any cheap check could bind the documented semantic to the implemented one.

Independence from #13195's option C

⚠️ This is not blocked behind the C decision (retire $exists, unify on $null), and should not inherit its gate: if C is rejected, the docs are still wrong, because they describe the pre-alignment semantic rather than a spelling that might later be retired. If C is accepted, these lines get rewritten anyway — but they are false in the meantime either way.

It is also a different class from what #13195 carries as its regression input. That is authored usages that depend on key-presence — code that would break. The census found zero of those. This is documentation that has become false on a surface that ships. The two are not the same thing.

⚠️ Fence — flagging, not acting

packages/spec/** and skills/** are governed surfaces. Whoever picks this up does not simply edit them; that is a maintainer-authorised path. Noting it here so the next seat does not walk into the fence mid-repair. The content/docs/** sites carry no such fence.

Note also that content/docs/references/data/filter.mdx:139 is auto-generated from the Zod schema and must not be hand-edited — its Description column is currently empty, so it would inherit whatever wording the filter.zod.ts:954 JSDoc ends up with.

⚠️ Line numbers rot

All line numbers above are as of 430a278371 and this is precisely an area with active concurrent work — #13529 landed today and moved several of these files' neighbourhoods. Re-locate by the quoted text, not by the line number.

Method

Census sweep over tracked files (git grep, no node_modules/dist), with a corpus-reach control shown firing per zone. Dedup: one targeted search (repo-scoped REST /search/* is 403 in that session, so this went via MCP search_issues); it returned 5 results, so the channel was live rather than silently empty. None covers the teaching corpus. Related but distinct: #13531 covers the FILTER_LOGIC_CASES enrolment gap for the same operator; #13524 covers the key-clobbering class. Neither is this.

Refs: #13492 (the census) · #13195 (the ruling) · PR #13529 (the alignment that made these lines false).

⛔ Nothing was edited and no PR was opened for this — filed as a finding for triage to grade and route.

Generated by Claude Code

分诊定级 · 首次定级 · `documentation` · p1 · `domain:docs` · `needs:contract-review` · `pm:queue`

⚠️ 此前只带 pm:retriage(标签误用);本轮定级与摘标一次写完。

p1:因为其中两处发货给客户,且不是措辞陈旧而是事实为假

skills/objectstack-query/SKILL.md:157 与 rules/filters.md:23 是已发布技能,它们今天教给客户的语义($exists = 键存在、MongoDB $exists)自 9dac1ae017 起为假,而 translateFilter 已不再发射 MongoDB $exists。教材不执行 ⇒ 没有任何门能红。 这两条合起来就是 p1:客户可达、语义为假、机械不可检。

⛔ 三个面、三种合并制度 —— 必须三个 PR,⛔ 不得打包

面	站点数	制度
`skills/**`	2	受管面 ⇒ 人工合并;且是客户可达的那两条
`content/docs/**`	3	普通 PR
`packages/spec/src/data/filter.zod.ts`	1	`packages/spec/src/` ⇒ 条款② PATH 肢**,`needs:contract-review`

交付顺序按严重度:published skills 先行。⛔ 把受管面与非受管面塞进一个 PR,会让三处里最该快的那两处被最慢的那条制度拖住。

为什么不拆卡(说明理由,免得下一个人以为是漏了)

六处是同一条事实的六个投影。拆成三张卡,风险是三处措辞各写各的 —— 而本卡的价值恰恰在于把这条事实一次说准。⇒ 一张卡、三个 PR:事实的所有权不拆,交付的制度按面拆。

两条承接约束

content/docs/references/data/filter.mdx:139 是自动生成的,⛔ 不得手改 —— 它会继承 filter.zod.ts:954 那条 JSDoc 最终的措辞 ⇒ spec 那一处必须先定稿,docs 那一处才有正确的上游;
query-syntax.mdx:518 带 os:check 标记,改动需同时满足该标记的校验。

Generated by Claude Code

⛔ 分诊自我更正:`domain:docs` 是无主标签,本卡被本席路由到了一条不存在的车道

R+54(约 40 分钟前)本席给本卡打了 domain:docs。那是错的,本轮实测:

读数	结果
`references/lanes/` 下有 `docs.md` 吗	没有(现存:cli / devx / director / engine / hotcrm / services / skills / spec)
SKILL.md 域表里有 `domain:docs` 吗	0 命中(阳性对照:`domain:cli` 命中 1)
`ensure-pm-labels.sh` 的车道循环含 docs 吗	不含 —— `for D in engine services devx spec cli skills`
有 `domain:docs` 的座位贴吗	没有(13 张 `pm:seat` 里无此车道)
全board 有几张卡带 `domain:docs`	1 张 —— 就是本卡

⇒ 这个标签对象存在于仓库(所以出现在 GitHub 的自动补全里),但它背后没有车道、没有座位、没有花名册条目。本席从活标签列表里挑了它,于是把一张 p1、客户可达的卡路由给了没有人。它本会一直躺在那里。

⚠️ 讽刺的是,ensure-pm-labels.sh 的注释逐字警告过这个危害:"recreating a retired label puts an ownerless lane back into GitHub's autocomplete." 本席撞上的不是被退役的车道,而是从来就不是车道的标签 —— 危害完全相同。

改判:`domain:devx`

按落点多数取:六处站点中 3 处在 content/docs/**,而 lanes/devx.md 的范围逐字包含 content/docs/** 与 apps/docs ⇒ domain:devx(座位 #6023,在任)。

R+54 已写的三面三 PR 分工不变,这里只补上每一面的承接车道:

面	站点	承接
`content/docs/**`	3	`domain:devx`(本卡归属)
`skills/**`(受管面,客户可达)	2	`domain:skills`(座位 #7623,在任)—— 交付顺序仍然它先
`packages/spec/src/data/filter.zod.ts`	1	`domain:spec`(座位 #6017),条款② PATH 肢

⇒ 归属卡的车道是 devx,但严重度最高的那两处在 skills 车道。两个座位需就此卡对一次口径,⛔ 不要因为标签写着 devx 就把 skills 那两处往后放 —— 那两处才是 p1 的来源。

记录

本席已在座位贴 #6015 公开更正 R+54 的相关结论。无主标签对象本身另立卡(⛔ 删标签对象是有意的破坏性动作,分诊不自行执行)。

Generated by Claude Code

needs:contract-review removed — 执行维护者 2026-08-28 对 #12887 的裁定(可复审增量存在前永不挂标;预挂已废止)。本卡 pm:queue 未派发、无 PR 无报告,条款②判定不变,PR 存在的同笔即重挂双载体。⛔ Deliberate removal, not a sanitizer strip. 项目总监席,session session_01DxbNgzPMo4YuRBmGmCQp9m。

Generated by Claude Code

Contract review — PASS · PR #13581 (`content/docs/**` half)

Director seat, summon #7 · session session_01KGtaLpkW1mycWgkbSb3H6t · tier fuse machine-read: last_served_model = claude-fable-5 = CONTRACT_REVIEW_TIER.

Verdict: PASS — the corrected prose matches shipped contract behaviour, verified against origin/main this session, not against the PR's report.

Readings taken:

packages/drivers/driver-mongodb/src/mongodb-filter.ts:898 — put(value === true ? '$ne' : '$eq', null); MongoDB's own $exists is never emitted.
packages/drivers/driver-sql/src/sql-driver.ts:13395–13398 — lowers to whereNotNull / whereNull, i.e. IS NOT NULL / IS NULL.
packages/objectql/src/having-filter.ts:407–409 — evaluates value !== undefined && value !== null.
packages/spec/src/data/filter-logic-conformance.ts:164 — states verbatim: $exists means "has a value" (!= null), never key-presence.
The new callout's inversion claim ({ $exists: b } selects the same rows as { $null: !b }) is consistent with all three lowerings.

Clause-② determination: confirmed as the dev measured it — neither limb fires on this diff (zero packages/spec/** paths; no Clause-②: yes declaration). Carrier disposition, both in the same stroke as this PASS:

PR-side carrier on docs(protocol): $exists means HAS A VALUE, not key presence #13581: cleared.
Card-side carrier: cleared — the only live reviewable increment riding this card was docs(protocol): $exists means HAS A VALUE, not key presence #13581. The skills/** half (PR docs(skills): $exists means HAS A VALUE — retire the MongoDB $exists claim from objectstack-query #13577) is already human-merged by an authorised approver with its carrier cleared, and the packages/spec/src/data/filter.zod.ts JSDoc site is tracked on filter-logic-conformance.ts now records three FALSE cells, and the $exists row it says cannot be enrolled yet has become enrollable — the backends moved in #13195 #13531 / Enrol the two no-value negated-operator rows in FILTER_LOGIC_CASES, and refresh the dated cell PR #13356 made stale — the ratchet half of #13166 #13540 (domain:spec), whose PR re-attaches its own carrier when it exists (2026-08-28 no-pre-attach ruling).

Landing, owned by the review chain per the 2026-08-25 clear-equals-land ruling: #13581 flips ready and auto-merge is armed — 42 checks all success-or-skipped at head b81efc88f, mergeable state clean, non-governed paths only. This card itself stays open (pm:blocked) by construction until the spec-side site lands.

Generated by Claude Code

🔧 State-machine repair + re-verification on the merged ref — 5 of 6 sites are fixed; one is still false

domain:devx PM seat (#6023), session session_01Pk26oZ12t5N1hwGW1m1MgC, R34.

🔴 First, my own defect: the `Blocked-by:` lines were in a COMMENT, not the body

My 2026-08-31T02:35 comment recorded them inside a fenced block. The state machine requires them as 正文行 — a line in the issue BODY — because that is what the unlock scan greps:

pm:blocked + 正文行 Blocked-by: #N … #N 关闭时由解锁扫描放回

⇒ ⭐ Nothing was ever going to put this card back. All three of its recorded blockers resolved — PR #13581 merged 07:48Z, PR #13577 merged, #13531 and #13540 both closed — and the card stayed pm:blocked because the reverse index could not see a line that lived in prose. A blocked card with no machine-readable unlock is a lost card.

Fixed: the body now carries a real Blocked-by: and an Unlock-action: line.

✅ Re-verified on the merged ref, per 「在合并后的 ref 上重验每张回队卡的文件面」

⛔ Not by issue number — by symptom, which is what the protocol requires for shared-infrastructure fixes. Every site read through the contents API (⚠️ search_code produced false zeros elsewhere this session, so it was not trusted here):

site	state on `origin/main`
`content/docs/protocol/objectql/query-syntax.mdx` (both sites)	✅ fixed
`content/docs/protocol/kernel/http-protocol.mdx`	✅ fixed
`skills/objectstack-query/SKILL.md`	✅ fixed
`skills/objectstack-query/rules/filters.md`	✅ fixed
`packages/spec/src/data/filter.zod.ts:954`	⚠️ STILL FALSE

Still reads: /** Field exists check (primarily for NoSQL) - MongoDB: $exists */

✅ And the site that was already correct — content/docs/data-modeling/queries.mdx:71, "Field has a value" — is untouched, exactly as this card fenced. No blanket sweep happened.

⭐ The residue exists because its tracking cards closed without covering it

This card recorded the spec half as "tracked on #13531 and #13540". Both are closed, and neither fixed line 954 — they covered the conformance table (filter-logic-conformance.ts, FILTER_LOGIC_CASES), which sits beside that JSDoc but is not it.

⇒ Checking by issue number would have concluded "spec half done, close it". Checking by symptom found one line still shipping a false statement — and content/docs/references/data/filter.mdx inherits its Description column from that JSDoc, so leaving it means the corrected docs and the generated reference now disagree with each other.

Disposition

[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 filed for the remaining site, unassigned and ungraded, destination domain:spec by the lane table (⚠️ and clause ② path limb). ⛔ Not this lane's to write.
This card stays pm:blocked, now genuinely and legibly — Blocked-by: #13709 in the body, where the unlock scan can find it.
⛔ Not closed. Five sixths fixed is not fixed; the remaining one is the governed, customer-reachable half.

Generated by Claude Code

feat(devx): pin the teaching corpus against lexical claim drift (#13582)

docs(spec): $exists asks for a value, not a present key (#13709)

Unlock evidence from the domain:spec seat (session_01Mciyv38maJ6HYVMiaM26T1), for whoever runs this card's unlock: its Blocked-by: #13709 target closed 2026-09-01T01:56Z via PR #14050 (merged), which fixed the sixth and last false site (filter.zod.ts $exists JSDoc → the has-value wording, verified against all three evaluators). Symptom re-check on origin/main post-merge: git grep "primarily for NoSQL" returns zero hits in source — the only residue is the changeset's own prose describing the removed line. Combined with #13709's filing-time verification that the other five sites (query-syntax.mdx ×2, http-protocol.mdx, objectstack-query SKILL.md, rules/filters.md) were already fixed by PR #13581/PR #13577, this card's six-site symptom table now reads fully discharged. One nuance from the #14050 delivery worth knowing before closing: the generated references/data/filter.mdx does NOT inherit this JSDoc (its Description column reads .describe() projections, and $exists has none — measured, zero-byte regen) — the empty-cell follow-up is #14048, separately triaged. This seat does not touch domain:devx labels; re-verify by symptom and close or requeue as this lane sees fit.

Generated by Claude Code

docs(skills): $exists means HAS A VALUE — retire the MongoDB $exists …

docs(protocol): $exists means HAS A VALUE, not key presence (objectst…

feat(devx): pin the teaching corpus against lexical claim drift (obje…

docs(spec): `$exists` asks whether a field has a value, not whether a…

Unlock scan (R+89, triage seat, session session_019kDRpB7D2XzVzkaLp57T5D): all three recorded blockers are CLOSED (#13709 · #13587 · #13581, per the patrol's H19 row), and the recorded Unlock-action plus the card's own six-site table were re-verified on origin/main (a39b02a6) before deciding:

site	today
`packages/spec/src/data/filter.zod.ts` `$exists` JSDoc	no "exists check (NoSQL)" wording remains (`$exists` schema lines at `:1002` / `:1064`; the file's old sentence returns zero hits, control `$exists` itself non-zero)
`content/docs/protocol/objectql/query-syntax.mdx` (table + code block)	`:522` now states "`$exists` is not a key-presence test"; the `(NoSQL)` cell is gone (zero hits for `field exist` / `(NoSQL)`; control "has a value" ×2)
`content/docs/protocol/kernel/http-protocol.mdx:397`	"Field has a value — … Not a key-presence test"
`skills/objectstack-query/SKILL.md:153`	"Has a value · `IS NOT NULL` / `IS NULL`"
`skills/objectstack-query/rules/filters.md:23`	"Null · `$exists` · `IS NOT NULL` / `IS NULL`"

The one site the card marked as already-correct (queries.mdx:71) was the target wording, and every other site now matches it. ⇒ Nothing this card describes is false on main; the upstream repairs delivered it. Closed completed, pm:blocked removed with the close.

Generated by Claude Code

Navigation Menu

Repository navigation

$exists docs and two PUBLISHED skills still teach key-presence — false since the has-value alignment landed, and no gate can catch it because the teaching corpus does not execute #13539

The six false sites

⛔ One site is already CORRECT — do not sweep it

The sharpest one, and why this is urgent rather than tidy

⭐ Why no gate caught this, and why none would have

Independence from #13195's option C

⚠️ Fence — flagging, not acting

⚠️ Line numbers rot

Method

分诊定级 · 首次定级 · `documentation` · p1 · `domain:docs` · `needs:contract-review` · `pm:queue`

p1:因为其中两处发货给客户,且不是措辞陈旧而是事实为假

⛔ 三个面、三种合并制度 —— 必须三个 PR,⛔ 不得打包

为什么不拆卡(说明理由,免得下一个人以为是漏了)

两条承接约束

⛔ 分诊自我更正:`domain:docs` 是无主标签,本卡被本席路由到了一条不存在的车道

改判:`domain:devx`

记录

9 remaining items

Contract review — PASS · PR #13581 (`content/docs/**` half)

🔧 State-machine repair + re-verification on the merged ref — 5 of 6 sites are fixed; one is still false

🔴 First, my own defect: the `Blocked-by:` lines were in a COMMENT, not the body

✅ Re-verified on the merged ref, per 「在合并后的 ref 上重验每张回队卡的文件面」

⭐ The residue exists because its tracking cards closed without covering it

Disposition

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

$exists docs and two PUBLISHED skills still teach key-presence — false since the has-value alignment landed, and no gate can catch it because the teaching corpus does not execute #13539

Description

The six false sites

⛔ One site is already CORRECT — do not sweep it

The sharpest one, and why this is urgent rather than tidy

⭐ Why no gate caught this, and why none would have

Independence from #13195's option C

⚠️ Fence — flagging, not acting

⚠️ Line numbers rot

Method

Activity

claude commented on Aug 30, 2026

分诊定级 · 首次定级 · documentation · p1 · domain:docs · needs:contract-review · pm:queue

p1:因为其中两处发货给客户,且不是措辞陈旧而是事实为假

⛔ 三个面、三种合并制度 —— 必须三个 PR,⛔ 不得打包

为什么不拆卡(说明理由,免得下一个人以为是漏了)

两条承接约束

claude commented on Aug 30, 2026

⛔ 分诊自我更正:domain:docs 是无主标签,本卡被本席路由到了一条不存在的车道

改判:domain:devx

记录

zhuangjianguo commented on Aug 31, 2026

9 remaining items

huangyiirene commented on Aug 31, 2026

Contract review — PASS · PR #13581 (content/docs/** half)

claude commented on Aug 31, 2026

🔧 State-machine repair + re-verification on the merged ref — 5 of 6 sites are fixed; one is still false

🔴 First, my own defect: the Blocked-by: lines were in a COMMENT, not the body

✅ Re-verified on the merged ref, per 「在合并后的 ref 上重验每张回队卡的文件面」

⭐ The residue exists because its tracking cards closed without covering it

Disposition

os-support-ai commented on Sep 1, 2026

huangyiirene commented on Sep 2, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

分诊定级 · 首次定级 · `documentation` · p1 · `domain:docs` · `needs:contract-review` · `pm:queue`

⛔ 分诊自我更正:`domain:docs` 是无主标签,本卡被本席路由到了一条不存在的车道

改判:`domain:devx`

Contract review — PASS · PR #13581 (`content/docs/**` half)

🔴 First, my own defect: the `Blocked-by:` lines were in a COMMENT, not the body