[PM seat] domain:drivers — 🟢 huangyiirene

[claude]

本贴是 domain:drivers 座位的唯一权威登记。 座位贴协议(维护者 2026-08-06 批准):索引 = label:pm:seat,总入口 #4604(指针页)。

单写手规则:只有在任座位 PM 编辑本贴正文;接管/移交 = 改正文 + 一条审计评论。空缺争用:动手前重拉正文、审计评论时间戳先到先得、写后回读。活性判定惰性(会话座位查最近产出评论,>24h 无产出可回收)。

范围

packages/drivers/driver-* (driver-memory / driver-mongodb / driver-sql / driver-sqlite-wasm / driver-turso)

当前 PM(会话或 Routine ID)

🟢 In office — GitHub huangyiirene · session session_013ZU38vNzLXCscgdFZjKd53 · took over 2026-08-11 ~04:2xZ via /pm-dispatch drivers. Predecessor: os-zhuang(08-08→08-11,16 PRs)。Patrol tick armed continuously(in-flight ≤45 min / standby 60–70 min;本会话名下恰好一条 patrol pending)。

说明

⚡ #5499 冻结已全面解除(2026-08-11 两道裁决,均记录在 #5499):

• mongodb:comment 5249019855「5499 解冻 mongodb 相关开发,继续全速推进」;
• memory:comment 5252526378「driver-memory 也解冻,继续清剩下两行 DEBT」。
• ⇒ 常设分诊规则(driver 族新单挂 hold 引 [裁决] driver-memory / driver-mongodb 投入冻结 —— 维护者 2026-08-05 口径（跨单锚点） #5499)已退役;全族新单照常入 pm:queue。[裁决] driver-memory / driver-mongodb 投入冻结 —— 维护者 2026-08-05 口径（跨单锚点） #5499 保留为冻结期沿革锚点,不再门控任何东西。

⚠️ 「7498 插队」= #7398(08-10 号码笔误,已消耗);真实 #7498 属 devx 车道

Ledger(本任期,2026-08-11)

🏁 DEBT 账本清零 —— check:driver-conformance 报 40 covered cell(s), 0 in the DEBT ledger, 0 exempt(5 drivers × 8 case-sets 全 ok;LEDGER = [] 在合并 ref 9bf4dd0 上复核仍空)。本任期开局是 36 covered / 4 DEBT。

解冻批(mongodb)COMPLETE:5 PR 全 MERGED,零 REWORK —— #7528(#5346+#5376)、#7550(#6850+#6814ᵐ)、#7579(#7177,含棘轮补丁轮)、#7590(#7580 dateGranularity,维护者直派)、#7625(#6682ᵐ)。mongodb 侧:拒收面补齐、结构化 GroupByNode + count/count_distinct 对齐、dateGranularity 五粒度 lowering、$contains 去折叠、两 conformance cell enroll。

memory DEBT 清账 COMPLETE —— PR #7723 MERGED 69fde55(19:49Z),双 Fixes 关闭 #6814 + #6682(已核验 closed/completed)。作用面:memory-driver.ts count_distinct arm(两面)+ 九处 'i' 折叠站点三簇、memory-analytics.ts($addToSet 排空后取 size)、最后两行 DEBT 删除 + 两个 enroll 套件同 PR。dev 实测证伪了卡片/账本/aggregation-conformance.ts 三处共有的前提(「analytics 面已实现 count_distinct」——实际返回原始数组),该证伪已写进 PR 正文。

memory analytics 回显对齐 COMPLETE —— PR #7852 MERGED 9bf4dd0(21:30Z),Fixes #7117。generateSql() 的 LIKE 家族此前渲染成等值(无通配符),而 #7723 把执行面钉到 case-exact 后又多出大小写轴;两轴一并修:改用 GLOB + spec 共享 likePatternToGlobPattern(SQLite LIKE 无条件折 ASCII,LIKE 回显会在修好通配符的瞬间在大小写轴上与执行矛盾;选构造即选模式语言)。operatorToSql 名→名映射改为谓词构建表,|| '=' 落穿消失。检查形态按维护者 5234868334 的指令:回显 SQL 在真实 SQLite(sql.js,版本与仓库现有 pin 一致)上执行、与 query() 行集对比,44 例,八次反向验证方向先判后跑全中。dev 证伪了卡片前提 A(startsWith/endsWith 根本不可达,被 #5345 的门以 400 拒;真正落穿的是 in/notIn/set,$nin 回显返回查询行集的精确补集),并把 $exists 的 key-presence vs IS NOT NULL 差异声明为显式不等式而非藏起来。邻接项拆卡不夹带:#7853(generateSqlFromPipeline 的 JSON.stringify 把 RegExp 压成 {})、#7854(spec 侧陈述在 #7723 后过时)。

全部 dev 会话已归档、零孤儿。

In-flight(0)

串行队(已清空)

热文件 sql-driver.ts 本任期未触碰。

仍持有 / 待裁

• driver-memory 完全没有行级租户隔离(#3724 的未修姊妹面):多租户下静默不隔离 #6915(memory 租户隔离)—— 本席已递 needs-user-decision(评论 5259108859):锚点在 9bf4dd0 上全部复核成立;[裁决] driver-memory / driver-mongodb 投入冻结 —— 维护者 2026-08-05 口径（跨单锚点） #5499 解除改变了裁决前提 —— 冻结期只有路线 B(启动期硬失败,镜像 mongodb-tenancy-guard.ts)能通过,现在路线 A(真正实现行级隔离)也已许可。本席建议 B 并给出理由,但真正待裁的是:driver-memory 测试面替代:项目内测试后端迁到 sqlite :memory:(#5499 重启条件 · memory 半边,维护者 2026-08-06 立项) #5704 把 driver-memory 迁出测试后端之后,它到底是什么定位。一句话即可解锁,B 可立即开派。
• Turso remote:带 id/conflictKeys 但没匹配上的 upsert 仍会静默写入 NULL 自增号(#6944 拒绝闸门覆盖不到的那条腿) #7099 — 等 TursoDriver remote 面根本不生成自增号:RemoteTransport.create 自建 INSERT,auto_number 只是个 TEXT 列 #6944 disposition A(turso)。ADR-0053 D-E3：mongo 的遗留形态是永久态，但它既无 backfill 也无读修复——#4191 storage-form 轴最尖的一格无法闭合 #4220 — 环境(无真 mongod)+ A/B 迁移形状裁决(5249068981)。drivers(sql): 共享 canonical 修复表达式把「裸数字文本」当儒略日读,backfill 会把这个误读写进盘里(观察项) #6009(sql 观察 p3)、flaky: sql-driver-overlay-index-drift.test.ts 「stays silent on the SECOND boot, when the runtime ledger starts empty again」偶发判红 #6522(finding 已测量)。
• 候选 follow-up 待分诊:driver-memory's AnalyticsResult.sql loses a RegExp operand to {}, so the pipeline dump shows no pattern at all #7853、packages/spec prose still calls driver-memory's $contains fold open after #7723 closed it #7854(本任期 dev 新立,未认领);dogfood date-bucket parity server-free 适配;driver-mongodb tsconfig 不 typecheck 测试文件;ask-2 算子 / $in-on-array(Filter Protocol 层)。

Session-operational notes for a successor

• ⛔⛔ RE-ARM FAILURE ≠ RE-ARM DONE:MCP 断连 = 失败,ToolSearch 重载重试至真实 fire_at;无法 arm 轮末明说;arm 后回读。断连高频、重载即恢复。
• Notes-21 空字段签名(本任期六例)处方定稿:先核成员资格(added_to_merge_queue webhook 优先;push 探针是最强正读数 —— 向 PR 分支推一个空提交,被拒且理由为 merge queue ⇒ 在队且分支未被污染),未入队才 disable/enable 翻转一次。
• ⚠️ 队列条目会坏死:test(spec): register sys_api_key in SINGLE_RECORD_WRITE_ONLY (queue-blocking) #7803 入队 75 分钟从未铸出 gh-readonly-queue/main/pr--* 投机构建分支,而期间十余辆更晚的车都铸过 —— 判据是「已在队 + 长时间无投机分支」,处置是单次 disable/enable 翻转(有效)。
• ⚠️ check-run 镜像会卡死:test(spec): register sys_api_key in SINGLE_RECORD_WRITE_ONLY (queue-blocking) #7803 的 TypeScript Type Check 两次在 job 层全绿收尾后仍显示 in_progress(48+ 分钟 / ~15 分钟),把 mergeable_state 钉在 blocked。诊断法:actions_get(get_workflow_job) 读 job 层真值,别信 check-run 列表;处置:rerun_workflow_run(内容已知全绿时无风险)。
• ⚠️ runner TLS 抖动:fix(driver-memory): make the $contains family case-exact and answer count_distinct #7723 队列构建一次红在 Checkout repository —— server certificate verification failed,一个测试都没跑到。属基础设施,重排一次即可,已在 PR 留分诊痕。
• MERGED webhook best-effort(test(drivers): remove redundant object: key from mongodb/sql query test literals #7579 未送达)—— 落地判定恒以 origin/main 为准。反过来 subscribe_pr_activity 的 MERGED 推送在 fix(driver-memory): make the $contains family case-exact and answer count_distinct #7723/fix(driver-memory): render the analytics echo as the query it describes (#7117) #7852 两次都准时到,值得挂。
• 已知 flaky:metadata-fs watcher tests are merge-queue flaky, and #7208's 20s deadline hardening did NOT fix it — the event is suppressed, not late #7282 签名(metadata-fs watcher,expected [] to include 'anchor',机制 metadata-fs: selfWrites suppression is time-keyed, so a poll tick landing inside the 200 ms window can swallow an external edit #7335)PR CI 命中一次,rerun 一投即绿。
• 共享损害会撞车:Queue-blocking: #7769 opened update on sys_api_key without bulk — api-methods-batch-conformance fails every merge-queue (full-suite) build #7793(api-methods 全队列中毒,fix(platform-objects,plugin-auth): give API-key revoke/restore a working product route #7769 开了 update 未登记豁免)本席按共享损害规则立卡止血 PR test(spec): register sys_api_key in SINGLE_RECORD_WRITE_ONLY (queue-blocking) #7803;另一席并行立 main is red for every PR that touches packages/spec: #7769 gave sys_api_key update without bulk, and the conformance scan that catches it lives in a package #7769 never touched #7802 → PR fix(spec,platform-objects): register sys_api_key's deliberate no-batch decision, unblocking every spec PR (#7802) #7818 抢先落地,Queue-blocking: #7769 opened update on sys_api_key without bulk — api-methods-batch-conformance fails every merge-queue (full-suite) build #7793 被关为 duplicate、test(spec): register sys_api_key in SINGLE_RECORD_WRITE_ONLY (queue-blocking) #7803 关闭未合。中毒期间 main 停摆 ~2.5 小时、队列踢车十余辆。教训:立卡前先搜同签名的既有卡(search_issues 打测试名),能省一整条流水线。
• verify-don't-trust:本任期 dev 六次测量修正 PM/卡片,六次全对 —— 派发令三分区是机制保障。最新一例(driver-memory's analytics generateSql() renders the LIKE family with NO wildcards, so the echoed statement is an EQUALITY the pipeline never ran #7117)在卡片指的地方找到了比卡片更坏的缺陷。
• 环境:mongodb-memory-server 代理 403(mongodb 卡走翻译层);driver-memory 在-process,套件是真执行;配额断粮窗可拖慢云 dev,RUNNING 读数照常恢复,判死基线从最近活证据起算。
• skip-changeset 按 check name 最新 run 判定;标签写后回读;调度器晚 3–8 分钟;git fetch 会移动 main;跨席声明先例三次([PM seat] domain:spec-surface — 🟢 os-zhuang #6298×2、[PM seat] domain:devx — ⏳ vacant #6023)。

---

迁移注记:自 #4604 座位表迁移(2026-08-06);本贴即该座位唯一权威。

[os-zhuang]

Seat takeover audit — 2026-08-08.

• Incoming PM: GitHub os-zhuang, session session_01Hg9Pkg5nDedCRihRsdeCdX, started by the maintainer via /pm-dispatch drivers (maintainer-initiated handover entrance).
• Predecessor: session session_01WyvqvKMG6asi9aXjKE6xtx (registered 2026-08-05). Its final output was the ACCEPT pair at 2026-08-08 06:28Z — drivers: $regex 响亮拒收 + $icontains 各后端实现(#4706 裁决 B 案 · 驱动半边) #5702 (PR feat(drivers,objectql): $regex 响亮拒收 + SQL 族 $icontains 实现(#4706 裁决 B 案 · 驱动半边) #6549, MERGED 06:43Z) and drivers: distinct 的第三参在 memory 与 sql 上语义不同 —— 把 sql 那面的写法交给 memory,静默返回未过滤全集(运行期实测) #6320 (PR refactor(driver-sql)!: distinct 的第三参收成裸 FilterCondition (#6320) #6517, MERGED 03:38Z). Ledger verified empty at takeover: zero domain:drivers + pm:dispatched open issues, zero open PRs touching packages/drivers/**.
• Body edits in the same stroke as this comment: title/assignee/当前 PM triple updated; scope path refreshed to packages/drivers/driver-* (the packages moved from packages/plugins/), driver-turso listed explicitly.
• Standing instructions carried over unchanged: [裁决] driver-memory / driver-mongodb 投入冻结 —— 维护者 2026-08-05 口径（跨单锚点） #5499 driver-memory / driver-mongodb investment freeze (triage routing note kept verbatim in the body).

Read-back verification of the body edit passed (entities and session IDs intact).

---

Generated by Claude Code

[os-zhuang]

Lane status 2026-08-09 09:0xZ — 11 merged today, and now zero dispatchable cards

Posting this as one consolidated read because the lane has stopped for reasons that live in three different threads, and each is a decision someone else owns.

Merged today (11), each through an independent step-7 gate

#6518 · #6402 · #6394 · #6577 · #6543 · #6409 · #6401 · #6792 (bee5ffe66) · #5495 (7d1ff752b) · #6754 (51f2bb8c3) · #6943 (45e711ad2)

All eleven dev cards archived; zero idle containers.

The board now: 14 open, 0 dispatchable

State	Cards
pm:queue but blocked	#6944, #6915, #6231
finding, awaiting grading	#7011 (new), #6850, #6814, #6522
pm:on-hold	#6682, #6009, #5376, #5346, #4220
tracking	#5499, #4706

Three unblocks, in the order they would help

1. #6231 — a domain:* re-label, pending ~20h. All five landing sites are in packages/metadata (3) and packages/objectql (2); none is in packages/drivers/. This seat flagged it 2026-08-08 13:43Z and is still deferring under the single-producer rule — dispatching it from here would have this lane edit two other seats' surfaces, and engine.ts took several landings today. Re-anchor to domain:metadata + domain:engine-core (possibly a two-way split), or rule explicitly that it stays domain:drivers on subject-not-site grounds, in which case this seat takes it as filed. Its serial blocker cleared: #5574 closed completed. Anchors have drifted hard — engine.ts 3317 → 4124.

2. #6944 and #6915 — an A-or-B disposition ruling. Both are "the capability is absent" cards, and both offer the #3724 shape: A implement it, or B refuse loudly at startup instead of failing silently. A dev should not pick A or B unilaterally, so they are queued and not dispatched.

3. ⚠️ #6915 may be queued against this seat's own standing instruction — please confirm which wins. This post's body carries the maintainer's 2026-08-05 rule anchored on #5499: new driver-memory/driver-mongodb cards take pm:on-hold referencing #5499 and do not enter the queue, with an exception only for things affecting CI-green correctness. #6915 is a driver-memory card that triage promoted to pm:queue; a tenant-isolation absence is not a CI-correctness matter, so on my reading the standing rule should have kept it on hold. Either the promotion is an intended exception (say so and I will dispatch it under route B), or it should revert to pm:on-hold. I have not dispatched it either way. Same question applies to #6814 and #6850, which sit ungraded rather than on-hold.

A pattern worth a ruling eventually, not today

Four cards now sit on the frozen driver-memory/driver-mongodb faces that #5499 forbids fixing: #6915 (no row-level tenant isolation at all), #6916 (autonumber collisions land as silent duplicates — filed by another lane), #6814 and #6850. The freeze is doing its job on investment; what it does not have is a place to put findings that keep arriving about the frozen surface. Recording the accumulation, not forcing a decision.

Two things this lane learned today, kept for whoever sits here next

• A queue build can go red on a package the PR never touched. fix(driver-sql): re-seed a stale autonumber counter instead of burning a number per failed create (#5495) #6932's red was PR fix(spec,cli,runtime,service-datasource)!: one driver vocabulary — os start and os migrate stop disagreeing (#6345) #6910 rewriting packages/cli/src/utils/storage-driver.ts — the oracle the --database-driver mysql 与 --database-driver sqlite-wasm 在 flag 解析阶段被拒 —— oclif 的 options 白名单漏了两个能用的驱动 #6860 pin derives its expectations from. Check file overlap, git merge-tree, and whether a co-queued PR rewrites the failing test's subject before treating a red as yours (ci: merge queue 吞吐优化——Test Core / Dogfood 各 2→3 分片 + 队列失败自动分诊评论 #4859).
• packages/objectql/src/engine-unknown-option.test.ts broke two different mechanical sweeps today, in two different ways: its bypassTenantAudit casts are load-bearing ([finding][drivers] 27 dead { bypassTenantAudit: true } as any options-gate casts in driver-sql / driver-sqlite-wasm tests — unlike turso's, these sit on a REAL typed gate #6754), and it holds a deliberately-unequal engine.find('task', { object: 'person' } as any) rejection test that a "delete every object:" sweep destroys ([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). Read it before writing any repo-wide pattern.

Seat remains held (os-zhuang, session session_01Hg9Pkg5nDedCRihRsdeCdX); dropping to standby cadence until one of the three unblocks lands.

---

Generated by Claude Code

[os-zhuang]

All three blockers ruled — closing the loop on comment 5230719256

Maintainer, 2026-08-10 chat:

#6915 和 #7117 就按 #5499 挂 on-hold,#6231 派给 metadata 席

Executed:

Card	Was	Now	Record
#6915	pm:queue + domain:drivers	pm:on-hold + domain:drivers	comment 5234867646
#7117	pm:queue + domain:drivers	pm:on-hold + domain:drivers	comment 5234868334
#6231	pm:queue + domain:drivers	pm:queue + domain:metadata	comment 5234873318 — handed off with re-measured anchors

The #5499 standing rule in this post's body stands as written and was not relaxed. The divergence was that triage had queued two driver-memory cards against it; the ruling settles it in favour of the freeze. Neither finding is retracted — both are held, with their measurements and their #3724-shaped A/B questions intact for whenever the freeze is revisited.

#6231's handoff carries fresh measurements rather than the card's stale ones: all five sites re-verified on main @ 3e8e669c0, engine.ts drifted 3317 → 4135 (it moved again between 07:3xZ and 01:1xZ, so locate by content), the premise re-confirmed (query.object still has zero readers in any driver), and the serial blocker #5574 confirmed closed. The engine-unknown-option.test.ts:183 trap — a deliberately-unequal { object: 'person' } as any rejection test that a naive sweep would destroy — is called out there explicitly, along with the note that the same file broke a second, unrelated sweep in this lane on 08-09.

Lane state after the ruling

pm:queue + domain:drivers is now empty. Remaining: five finding awaiting triage grading (#7099, #7052, #6850, #6814, #6522), seven pm:on-hold (#7117, #6915, #6682, #6009, #5376, #5346, #4220), two tracking (#5499, #4706). target:v17 ∩ domain:drivers = zero.

Note for the next occupant: four of the on-hold cards (#7117, #6915, plus #6814 / #6850 once graded) sit on the frozen driver-memory / driver-mongodb faces, and #6916 was filed against the same surface by another lane. The freeze is doing its job on investment; findings about the frozen surface keep arriving regardless, and this post is where they accumulate.

Seat held (os-zhuang, session session_01Hg9Pkg5nDedCRihRsdeCdX), standby cadence.

---

Generated by Claude Code

[os-zhuang]

交接审计 — os-zhuang 卸任(2026-08-11 ~04:0xZ)

维护者 2026-08-11 指示「新项目经理会接手」。本评论仅作交接存档,状态以正文为准(座位贴单写手协议)。

交接点:in-flight ZERO。 无在派 dev 会话、无 open PR、无排队条目、无孤儿订阅、无未归档云卡。与 2026-08-08 接手时同样是干净的切点。

项	值
在任	2026-08-08 → 2026-08-11 04:0xZ
会话	session_01Hg9Pkg5nDedCRihRsdeCdX
任期落地	16 PRs MERGED(08-09 ×12,08-10 ×4),另 #6522 无 PR 关闭(测量结论)
末次实测	03:3xZ:domain:drivers 14 open · 队列空 · origin/main 211abdbd6
巡逻 tick	已解除武装(trig_01Tsk7vgcpb7jTLVhgfwrbR1 删除)——接任者自行 arm

08-11 是纯待命段:零派发、零落地,三条只读记录。 车道确实空(自 08-10 18:4xZ 起 14 open 一格未动),不是没找。三条记录分别是 #6522 的过期理由更正(5247751272)、#7177 的不派理由入卡(5248086029)、#7501 的落点判据(5248698620)。三者都不改标签、不重定级——定级是分诊的职权。

给接任者的三个最高优先提示(展开见正文):

1. ⚠️ 「7498 插队」= driver-sql: a declared operator on a multiple: true (JSON array) column silently answers wrong — $in/$eq always zero rows, $nin returns the rows it was asked to EXCLUDE #7398,不是今天新建的 pm-dispatch SKILL: require a four-lens analysis block on every needs-user-decision card #7498。 那是 08-10 的号码笔误,已消耗完毕并落地(PR fix(driver-sql): refuse scalar-comparison operators on JSON/multi-value columns instead of answering silently wrong #7415)。今天 02:58Z 真的出现了一张 pm-dispatch SKILL: require a four-lens analysis block on every needs-user-decision card #7498(devx 车道,pm-dispatch SKILL),两者无关,本座位未碰。前任还就此多问了维护者一句——答案本就写在座位卡上,属多余动作,记下来免得重演。
2. ⛔ 本车道内标签 ≠ 授权,直到 [drivers] driver-memory / driver-mongodb disagree with the SQL family on count_distinct — the two AGGREGATION_CASES cells #6409 left open #6814/[finding][drivers] driver-mongodb cannot take a structured GroupByNode at all — the object stringifies into a "[object Object]" $group._id #6850 的七秒批量翻转事件解决(已报 [裁决] driver-memory / driver-mongodb 投入冻结 —— 维护者 2026-08-05 口径（跨单锚点） #5499 5237773754)。[finding][drivers] The object-in-query sweep of #6231 was source-only — 10 more sites of the same shape live in driver test files #7177 是同类形状的新样本:分诊晋级理由本身站得住,但没有权衡站点所在的包——12 个站点里 11 个在 [裁决] driver-memory / driver-mongodb 投入冻结 —— 维护者 2026-08-05 口径（跨单锚点） #5499 冻结的 driver-mongodb。
3. ⛔⛔ re-arm 失败不等于 re-arm 完成。 08-10 19:45Z→00:34Z 座位空 armed 约 4h49m,起因是 MCP 断连时 send_later 连续失败而该轮静默收尾。代价为零仅仅因为当时 in-flight 为零。

本任期两条自我更正,一并留档:座位卡旧版的 #7177「9/10 frozen」应为 11/12;#7501 上 scale → precision 的别名一度看着像烟枪,查证后作废(那条属 CurrencyConfigSchema,与字段级别名表不冲突)。dev 在本任期内 10 次证伪 PM 声称的前提且每次都对——预先定价的派发事实按构造是 verify-don't-trust,这条请一并继承。

座位现为 🔵 空缺待接管。接任者按协议:改正文 + 一条审计评论,写后回读。

---

Generated by Claude Code