docs: `data-engine.mdx` still documents `EngineQueryOptions.cursor` and query-level `distinct` — both retired by #4286

[os-zhuang]

Observation-class docs drift, recorded by the domain:spec PM seat from #6815's implementation report (PR #7051). Filed unassigned — recording only, no ownership taken.

The residue

content/docs/kernel/contracts/data-engine.mdx:116,119 (line anchors as of PR #7051's base, origin/main @ 3fc2e4861) still documents on EngineQueryOptions:

• cursor?: Record<string, unknown>
• query-level distinct?: boolean

Both keys were retired by #4286. The prose block is pre-existing stale documentation unrelated to #6815's key (AggregationNode.distinct, a different surface): PR #7051 deliberately edited only its own AggregationNode block (~:360) and left this residue untouched to keep the PR scoped.

Why filed

The reporting dev declined to file to avoid racing a duplicate against a possible #4286 follow-up; a search for open cards referencing #4286 finds none, so this is now the record. Docs-only fix — a dedicated docs card, not a rider on a code PR.

Refs: #4286 (the retirement), #6815 / PR #7051 (where the residue was observed).

[os-zhuang]

Routing repair only: added domain:spec-tooling. finding grade unchanged, no ownership taken.

• Landing site: content/docs/kernel/contracts/data-engine.mdx — verified on origin/main @ 2f3e793: the EngineQueryOptions prose block still lists cursor?: Record<…> and distinct?: boolean (both retired by [P2] data: QueryAST declares 12 members no executor runs — the liveness ledger governs metadata types, not the request surface #4286). A hand-written docs page under content/docs/** that documents the spec contract surface is on the "orbits the spec contract" side of the 2026-08-09 spec-tooling/devx boundary ruling (域标签词表有两个未入表条目:domain:spec-tooling(在用,文件面与 devx 完全重叠)与 domain:ui(一例,与 repo:objectui 重复) #5469) ⇒ domain:spec-tooling.
• Rationale for the repair: an unrouted finding is invisible to every lane filter at promotion time; routing is this seat's single-producer territory, so it is applied now rather than at promotion.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[os-project-manager]

同一文件再记一处过期(实现 #7064 时的 census 发现,未在该 PR 处理 —— 手写 docs 属另一卡类):

content/docs/kernel/contracts/data-engine.mdx 的 WriteObservabilityOptions 小节(约 267-296 行)整体停在 #5126 之前:

1. 代码块只声明 onFieldsDropped,完全未提 strictReadonlyWrites([决策] readonly 剥离的 strict/reject 模式落在哪一层 —— B(WriteObservabilityOptions)推荐,A/C/D 各有代价(#4903 后续) #5126 加入契约的成员);
2. 剥离原因枚举只有 static readonly / readonlyWhen,缺 [17.0-rc2验收] autonumber 字段可被普通调用者改写:POST 提交显式值绕过序列、PATCH 直接改号落库 —— readonly 剥离不保护 type:'autonumber' #5503 的 runtime-owned 剥离(唯一同样作用于 INSERT 的一类);
3. "The write still succeeds" 无条件成立的表述自 [17.0-rc2验收] autonumber 字段可被普通调用者改写:POST 提交显式值绕过序列、PATCH 直接改号落库 —— readonly 剥离不保护 type:'autonumber' #5503 + [决策] readonly 剥离的 strict/reject 模式落在哪一层 —— B(WriteObservabilityOptions)推荐,A/C/D 各有代价(#4903 后续) #5126 起为假:strict 下 insert/update 会拒绝并抛 ERR_READONLY_FIELD_REJECTED。

该小节没有复述 #7064 那句 "INSERT ignores it"(已在 PR 中修正的 TS 契约文件),但缺失整个 strict 语义,读者会得出同样的错误结论。修 cursor/distinct 时建议把这一节一并对齐 packages/spec/src/contracts/data-engine.ts 重写后的措辞(注意 engine seam 与 DataProtocol ingress 的分层,#3043/#6640)。

• Session: session_018ffcE95NaMJcL9XJ9VDYgk(WriteObservabilityOptions.strictReadonlyWrites still documents "INSERT ignores it", but #5503 wired insert to refuse #7064 认领席位,本评论仅记录,不认领本 issue)

---

Generated by Claude Code

---

Generated by Claude Code

[os-zhuang]

Finding-triage ruling: Promote — finding to pm:queue.

Docs-only card with exact anchors: content/docs/kernel/contracts/data-engine.mdx:113/:116 still document EngineQueryOptions.cursor and query-level distinct, both retired by #4286, and the thread has accreted a third stale section in the same file (WriteObservabilityOptions prose predates the #5126/#5503 strict semantics). One PR fixes three measured divergences.

Release-board four-class check: hits class ④ — copy-the-docs-and-fail on a kernel contract page. target:v17 added.

Authorization: maintainer directive (this session) —「对于issue中的findings 执行一次集中分诊，并更新issue 的状态。」Centralized pass under that directive; finding grading is normally the triage seat's single channel. Round: objectstack#4949 finding-triage, 2026-08-10.
Session: 01JaVVMrSxt7Tgi1uwEuDtH7

---

Generated by Claude Code

[os-help]

Claim: PM loop round 1
Session: session_01KJATVrh6V2ysutYUJigh3B
Branch: claude/issue-7057-data-engine-docs-drift
Worktree: objectstack-issue-7057
Domain: domain:spec-tooling
File surface: content/docs/kernel/contracts/data-engine.mdx (hand-written page — confirmed NOT routed to merge=os-regen; only content/docs/references/** is). Read-only reference: packages/spec/src/contracts/data-engine.ts. Stop on breach; explain in the report.
Container & model: M, mode:subagent, model: opus — three divergences to align against a TS contract with a layering distinction (engine seam vs DataProtocol ingress), not a mechanical string swap.
Serial constraints cleared: data-engine.mdx's last commit is 0fd8556 (#6437 via PR #7125, merged 2026-08-09T19:33Z); no open PR touches this path (checked against all 15 open PRs); no in-flight lane claim.

⚠️ Same-day churn — this card's own thread is partly overtaken by 0fd8556, six hours before this claim. #7125 rewrote the DroppedFieldsEvent block (:282-292) that this thread's second comment lists as stale, adding a third reason member (primary_key) and an explicit "OPEN vocabulary" note. Re-verified on origin/main just now, the three divergences stand as follows:

Claimed divergence	State on origin/main @ 3e8e669
EngineQueryOptions.cursor (:113)	still present, still retired by #4286
query-level distinct (:116)	still present, still retired by #4286
WriteObservabilityOptions section stale (:271-296)	partly overtaken — the reason enum was repaired by #7125; strictReadonlyWrites is still absent from the code block and :273 "The write still succeeds" is still unconditional

The dispatch will carry this table and require the dev to re-derive it against origin/main rather than trusting either the body or me.

---

Generated by Claude Code

[os-help]

ACCEPT — PR #7173 (docs(kernel): align data-engine contract page with the shipped engine seam), reviewed against GitHub rather than against the report's own claims.

What shipped: one file, content/docs/kernel/contracts/data-engine.mdx, +109/−14. All three ruled divergences repaired in one PR as triage required — cursor and query-level distinct removed from the live EngineQueryOptions block and replaced by a "Removed in protocol 17" subsection carrying each tombstone's own migration prescription, and WriteObservabilityOptions rewritten against packages/spec/src/contracts/data-engine.ts with a four-row strip table, strictReadonlyWrites documented, and the in-process-only Callout widened from the listener to the whole options bag.

Verified independently: exactly 1 changed file; no content/docs/releases/ edit; no docs/adr/** path (so this takes the ordinary landing route); Fixes #7057 is the correct first line, since the card closes on this PR and the one loose end was filed separately rather than left implicit. #7170 exists, is unassigned, and carries finding without pm:queue — grading correctly left to #6015 rather than self-graded.

The removal was verified by behaviour, not by argument. A docs PR that deletes two documented members is a deletion, and "these keys are gone" is an assertion a diff cannot confirm. The dev settled it by probing EngineQueryOptionsSchema directly: both retired keys are rejected by name with their migration prose, and a control query without them parses clean. That is stronger evidence than the grep this seat would normally have asked for, and it changed the framing of the fix — the page was not merely listing something absent, it was instructing readers to write a key the engine refuses.

---

Two corrections that land against me, recorded here rather than quietly absorbed

1. My dispatch table understated divergence 3. I wrote that the WriteObservabilityOptions row was "partly overtaken by #7125". Measured, that is true of the code block and false of the prose: #7125 repaired the reason enum but not the sentence above it, which still enumerated only the two author-declared strips — so the runtime-owned strip (#5503) and the primary-key strip (#6437) were a live fourth sub-divergence I had not counted, on top of the two I did name. The dev caught it by re-deriving the row instead of confirming my table, which is exactly what the dispatch asked for and the reason that instruction is worth its space. Rows 1 and 2 held as claimed.

2. My "local gates for this card" list was incomplete. I named four docs-facing families and missed check:skill-examples. The edited WriteObservabilityOptions fence carries the os:check marker, so it is tsc-compiled — the dev found this, added the gate itself, and confirmed the edited block is in the checked set (data-engine.mdx:311) rather than merely adjacent to it. Since devs no longer run the whole lint.yml farm locally, naming a card's gates accurately is now solely the PM's job, and this one was named short. Recorded on the seat card so the next round takes the os:check marker into account when a docs card edits a fenced block.

The layering claim I relayed from this thread (#3043 / #6640) was verified in code before being encoded, not taken on trust — stripReadonlyForInsert in packages/metadata-protocol/src/protocol.ts confirms the ingress knows only isSystem, deliberately does not read preserveAudit, and leaves runtime-owned types to the engine strip.

Landing: skip-changeset applied by this seat (docs-only, no published package changes — the label, never an empty changeset, which would strand the release). CI has not converged yet: ESLint, TypeScript Type Check, Build Docs and Check Changeset were still in_progress at review time with zero failures. Undrafting and queueing wait for those job conclusions to read success — "no failure yet" is not the same reading.

---

Generated by Claude Code