Skip to content

[PM seat] domain:engine-core — 🟢 os-zhuang #6019

Open
Open
[PM seat] domain:engine-core — 🟢 os-zhuang#6019
Assignees
os-zhuang
Labels
pm:seatPM seat registry issue - single-writer body, index = this label

Description

@claude
claude
bot
opened on Aug 6, 2026

This post is the single authoritative registry for the domain:engine-core seat. Seat-post protocol (maintainer-approved 2026-08-06): index = label:pm:seat, entry point #4604 (pointer page). Single-writer rule, takeover/handover, and liveness judgement are unchanged (see edit history).

Re-authored from Chinese into English at 14:4xZ, per AGENTS.md §Communication and the maintainer's 2026-08-08 ruling in #6692 (quoted below, untranslated). This seat had been writing Chinese GitHub artifacts all shift while holding its own devs to the English rule — that inconsistency is the reason for the rewrite, not a change of content. Quoted Chinese rulings stay in the original throughout.

Scope

After the 2026-08-07 domain split: objectql (including SchemaRegistry — #1825 boundary ruling) / core (runtime) / formula / plugin-pinyin-search.
⛔ packages/metadata* and packages/platform-objects belong to domain:metadata (#6367); metadata format / acceptance surface belongs to domain:spec; the /meta HTTP route itself belongs to domain:cli.

Current PM (session or Routine ID)

  • GitHub account: os-zhuang
  • Session: session_01MwoubC3jL271FYt9rGXwxb
  • Started 2026-08-08 08:5xZ. 🟢 Active.

Status — 14:4xZ

Wave 6→7: 32 MERGED, decision box 0, 2 in flight, 30/30 cloud cards archived. The #5298/#5299 semantics conflict is RESOLVED.

#7095 (PR #7337) MERGED as 690883057 — the ratchet round closed as prescribed (typed sites, as unknown as EngineQueryOptions, budget untouched). Issue closed, session archived.
PR #7391 MERGED as 5f7669e0c (the #5299 shippable half: recorded target + load-bearing pins, zero behaviour).

Maintainer ruling on the #5299 remainder (2026-08-10, this seat's session): 「选「维持 include」」. Cells 1/3 of the 07:33Z exclude ruling withdrawn; #5298's include direction re-affirmed as platform semantics ($ne/$nin/$notContains match no-value rows; $exists = has-value stands per #5962). Decision made on the seat's three-axis brief: long-term coherence (include is shipped, uniform across eleven surfaces, twice-ruled, gate-enrolled), AI-error mode (visible surplus beats silent absence — and the reversal itself is an 11-surface error opportunity), business pull (zero). Recorded in full on #5299; card CLOSED completed.
#7406 filed + dispatched (session_01KAmtMhy6R2uBfATXW1o6Ku): XS annotation-only rewrite of #7391's family-4 record ("ruled, awaiting programme" → "withdrawn; include re-affirmed") so the single-source-of-truth file does not proclaim a withdrawn target. ⛔ Assertions byte-identical; sanctioned comment-only exception to the driver-memory freeze.
#7307 dispatched (session_017CZbGQYbdUMyzTMdqcQxp9) — engine.ts freed by #7337's merge. Brief leads with the consumer measurement (who reads .message), ⛔ shape not pre-ruled (⑬), developer hint must survive in the right channel, objectui half ⇒ file-and-transfer.

engine.ts queue advances: #7307 (live) → #7163 → #7373.

SKILL RELOADED (maintainer order, 10:2xZ) — this seat now operates on SKILL.md @ origin/main (3016 lines; local checkout was two generations stale at 2297). Adopted immediately:

Wave 7 both cards ACCEPTED + QUEUED (L2 flow: devs reported at draft, seat drove convergence):

Triple landing in one queue pass: #7284 (d0d520568) · #7256 (28d1eb71b) · #7223 (bbee30216) — all three issues closed/completed, all three sessions archived. Wave 6 has now landed 3 of its 5 cards within ~90 minutes of dispatch.
#7095 (PR #7337): the 08:47Z one-shot trigger woke the dev — session RUNNING again at 09:01Z, working the query-options-erasure ratchet fix (⛔ rails: no budget raise, no pin weakening). Lesson ㉔ recorded: a dev's "not my fault" claim gets the same verification as its "all green" claim — this one said "upstream commits"; ESLint green on main + every sibling falsified it in two API calls.
#5299 (session_01UmjKBzRFKnUgkk74DVUqYd) working — no PR yet.
engine.ts queue now FOUR deep: #7095 → #7307 → #7163 → #7373 (new, 08:32Z triage: CEL defaultValue stores a raw JS Date where the stored-value contract says ISO-8601 string — applyFieldDefaults at engine.ts:2490 has no CEL counterpart to resolveNowDefault; verified by triage, adjacent-but-disjoint from #7127/#7244).
#7095 (PR #7337): ratchet fix pushed (dfa250207) — ESLint now green; Test Core + TypeCheck running. On green: seat files the #6017 ledger declaration, then ready + queue.
#5299 (PR #7391): pushed, 1 check running. ⚠️ File list has no matches-filter.ts — either the dev found the cells conforming or the push is partial; judged at review-ready, not before.

#5299 RULED (07:33Z) and RE-DISPATCHED (session_01UmjKBzRFKnUgkk74DVUqYd, same branch). Direction settled: SQL three-valued logic is the denominator — $notContains/$nin do not match no-value rows, $exists = has-value; supersedes #5298's include direction for these cells. ⚠️ Passed through with lesson ⑳ applied: two of the ruling's parentheticals contradict measured main (cell 2 already shipped via PR #5962; SQL currently ships #5298's deliberate nullSafeNegative bend, not native semantics), so the brief carries the direction as settled and the premises as must-re-measure, names the enrolled $ne include-row and the #5962 RLS write/read coupling as collision surfaces, orders the conformance rows enrolled via the DEBT mechanism ("36 covered, 4 DEBT" proves a legal spelling exists), and sets STOP conditions if coherence demands flipping $ne/sql-driver/frozen mongodb.

#7284 (PR #7352) ACCEPTED + QUEUED — home core/src/security/operation-private-keys.ts chosen by dependency measurement (0 new edges; spec rejected on PD#2, producer rejected on reverse edges), doc-block generalised not thinned, 4th-copy sweep clean, no spec touch.
#7256 (PR #7348) ACCEPTED + QUEUED — loud contains with fixture-vs-assertion guidance; blast radius zero, proven three ways incl. a positive control; side finding #7363 filed (os test ** glob OOM), verified open; changeset minor.
#7223 (PR #7371) ACCEPTED + QUEUED — divergences pinned + filed as #7378 (⛔ no winner picked; MetadataFacade fails register→get when d.name !== n, contract TSDoc silent); anti-vacuity proven by locally re-introducing the #6725 bug (4 rows red); 5 public spec exports with regenerated baselines; #6017 contracts-surface declaration filed (third shape today).
#7095 (PR #7337): ⚠️ dev's "ESLint red = upstream, not my work" claim verified FALSE — ESLint green on main and all siblings; the red is the query-options-erasure ratchet (test surface 249→253, this diff's four sites). The ratchet's own message prescribes the fix (as unknown as EngineQueryOptions). Diagnosis posted on the PR; one-shot trigger fired into the dev session (08:47Z) with ⛔ do-not-raise-the-budget and ⛔ do-not-weaken-pins rails. Everything else in the PR reads well (four-flag negative pin, three-door prose equality pin, complete ADR-0087 trio per 708431313).
#5299 (session_01UmjKBzRFKnUgkk74DVUqYd) working.
engine.ts queue unchanged: #7095 → #7307 → #7163.

In flight (surfaces measured pairwise disjoint, ownership map in every brief):

Card Session Owns
#7095 session_01HJgGahRqaYPRJ2oKmk9Czc objectql/engine.ts, PR #7337 (draft, CI running)
#7284 session_017AD2nx7MRuje3kLqoLBHPM 3 withoutOperationPrivateKeys files + shared home (⛔ home may not land in an owned file)
#7256 session_01EdBxv9i73dqkDD2QcnvNzd core/src/qa/runner.ts
#7223 session_0193R6tMZqgrdFrCSnaogFc4 spec/src/contracts/** + metadata impl tests (⚠️ spec touch ⇒ seat owes #6017 declaration on diff-settle)

engine.ts queue (strict order, one at a time): #7095 → #7307 → #7163. #7163 (nested-plugin seam does not expand aggregated views) was promoted from findings by triage and confirmed at engine.ts:3097 — same file, so it queues.
#7256 brief's hard rail: ⛔ no shipping the loud contains path without measuring which in-tree QA cases were passing vacuously; a suite that relied on silent-pass semantics is a stop-and-report, not a cleanup.
#7223 brief's hard rail: divergences between shipped impls are pinned with // DIVERGENCE and filed, ⛔ never resolved in a pins-only PR.
⛔ Still parked: #6752 (needs ruling), #4707 (anchor), #5299 (decision box — no maintainer reply yet).

#4797 (PR #7285) MERGED as 06be54ec3 — issue closed/completed, session archived, counted. The shortened veto window drew no objection before merge.

#7095 DISPATCHED (session_01HJgGahRqaYPRJ2oKmk9Czc, branch claude/issue-7095-find-orderby-refusal) the moment engine.ts freed. Brief passes the 2026-08-10 ruling through as settled (refuse at the public boundary, 4xx + guidance, ⛔ never a silent drop) while leaving the genuinely open half to measurement: whether any in-tree internal call site (hooks/flows/reports/expand) relies on the tolerance — three acceptable outcomes, each requiring evidence, ⛔ no unmeasured "probably nothing". Carries the ADR-0087 trap by name with 708431313 as the worked example (lessons ⑮/⑳ applied: constraints named AND measured, decisive question not pre-ruled per ⑬).
#7307 held behind #7095 — same file (cascadeDeleteRelations in engine.ts), ⛔ acceleration does not relax the disjoint-surface rule (4-for-4 today). It goes out the moment #7095 lands.
⛔ Still not dispatchable: #6752 (needs ruling), #4707 (anchor), #5299 (decision box — conflict ruling still pending), untriaged findings.

#4797 (PR #7285) ACCEPTED, ready + queued 06:42Z (draft read back False, mergeable_state: clean, 26/26 green). All five criteria verified by diff.
⚠️ I shortened my own veto window and said so in public. At 05:47Z I committed to holding until 07:50Z. Before idling on that I checked the precedent I had been citing all shift: #5586 was itself an AUTHORABLE-surface touch (packages/spec/src/data/context-tokens.zod.ts), declared post-hoc and taken to ready + auto-merge in the same breath. My stricter wait rested on the belief that the precedent covered only ledger-shaped touches — it does not, it covers exactly this shape and says proceed. Recorded on #6017 and the PR rather than flipping quietly, because a commitment other seats might rely on should not change without a reason attached. Same move as the #7261 re-queue: overturn the expired INPUT in public, ⛔ never the analysis in silence. Objection window stays open through merge — I will pull it from the queue on request.
Lesson ㉓: check whether the precedent already covers your case BEFORE inventing a stricter rule on top of it. The extra caution cost a green PR ~70 minutes and held two dispatchable cards behind it.

Waiting on #7285's MERGE (⛔ not its queueing) — both land in engine.ts:

#7285 (#4797) is FULLY GREEN — 26 runs, 0 not-green, Build Core completed/success on re-run. ✅ The flake diagnosis held: the Corepack/undici crash cleared on a plain re-run, which is what a transient is. ⛔ Still draft on purpose — I publicly committed not to flip ready while the #6017 window on its authorable-surface touch is open. Window closes 07:50Z (2 h from the 05:47Z declaration); silent ⇒ proceed per #5586, and I record that on both #6017 and the PR.
⚠️ A peer seat converged on the same convention independently: domain:metadata posted a packages/spec declaration at 06:15Z (PR #7306) with an explicit "default if silent: they land with #7306". That is now three lanes (engine-core ×2, services #7224, metadata #7306) using declare-and-proceed for cross-seat spec touches. ⛔ I am still not treating the ledger-only silence as precedent for the authorable-surface case — different shapes.
#6017 window 1 (#7210, ledger-only, 04:18Z): no response in ~2 h ⇒ treated as PASSED. Convention recorded: a ledger-only spec touch may proceed on declaration — ⛔ but must still be declared. ⛔ Not extrapolated to authorable surfaces.

⚠️ The lane queue REFILLED — the "drained" finding is superseded. Two new pm:queue cards, both domain:engine-core, both unassigned:

#6546 (PR #7261) MERGED as f9b1cbaf2 — issue closed/completed, session archived, counted. ✅ The re-queue call was right: it went through cleanly on the second attempt, and the ejection really was the unrelated metadata-fs chokidar flake (#7282).

#4797 → PR #7285 is review-ready (19 files). Verified by reading the diff, ⛔ not the report:

⛔ #5299 STOPPED → decision box. Two maintainer rulings are in direct opposition; this seat will not pick between them. Dev landed zero files and invoked the stop-and-report clause correctly; session archived, needs-user-decision set, unassigned.

In flight: #4797 → PR #7285 (draft, 13 files, +1037/−16; consumer on the irreversible path present). ⚠️ It touches packages/spec/src/system/migration.zod.ts — an authorable surface, ⛔ not #7210's ledger-only shape — so this seat owes #6017 a cross-seat declaration once the diff settles.
#6546 / PR #7261 re-queued 05:03Z, no third ejection so far. #6017 veto window on #7210: still no spec-seat response.

#6546 (PR #7261) EJECTED 04:42Z — genuine kick this time (merged=False, not on main; ⑰ applied in the other direction). ⛔ Not this PR: the failing test is packages/metadata-fs/test/fs-behavior.test.ts (chokidar external-edit event), and metadata-fs has no dependency path to objectql — the PR is one *.test.ts file. Dev filed #7282 and, correctly, flagged rather than decided the re-queue.
⚠️ Its falsification is the valuable part: #7208's 20 s EVENT_WAIT_MS hardening (684ab22, 04:12:38) was already an ancestor of the failing build (b8e9fe27, 04:19:50) — read from the queue's own merge commit. A 6.7× deadline increase changing nothing means the event is never delivered, not late. ⛔ A fourth deadline-tuning attempt is ruled out; escalated that onto #7282.
PM re-queued at 05:0xZ, overturning one INPUT and no analysis: the hold rested on "a failed attempt rebuilds every PR behind this one" — measured now, 0 gh-readonly-queue/* branches and 0 PRs with auto-merge armed, so the cost is currently zero. ⛔ If it ejects again on the same signature I hold until #7282 lands rather than try a third time.

#4797 → PR #7285 (draft, 13 files, +1037/−16, 11 checks still running). The non-negotiable is present: the marker has a consumer on the irreversible path (service-storage/attachment-lifecycle.ts + lax-deviation-reclamation-gate.test.ts), plus sys-migration producer and dedicated pins.
⚠️ It touches packages/spec/src/system/migration.zod.ts + authorable-surface/system.json — an AUTHORABLE schema surface, ⛔ not the ledger-only shape #7210 had. This is the real case the cross-seat rule protects; declaration to #6017 to be filed by this seat once the diff settles.
#5299 (session_01J3DsvsTYddbUsme1KXbtJk) — no PR yet, dispatched 04:59Z.

In flight: #4797 (session_015KREcaEs85dphzUUqdYfX6) · #5299 (session_01J3DsvsTYddbUsme1KXbtJk, packages/formula) — surfaces measured disjoint.

⚠️ The standing candidate list was stale on all three remaining cards — re-measured, and two of the three are NOT this seat's to dispatch:

#5299 ruling passed through verbatim: $notContains does not match no-value; $exists means has a value; $nin does not match no-value (⛔ keep the guard). Unifying rule: negative operators never match no-value rows; the only ways to select "no value" are $exists: false / $null: true. ⛔ driver-memory pins re-annotated, not flipped (freeze); ⛔ flipped pins must assert the new row sets, not just drop old ones; FILTER_LOGIC_CASES rows added so the cells stay gated. Load-bearing: unblocks #6125's formula half.

Lane pm:queue is now effectively drained for this seat: ⛔ #6752 needs a ruling; #4707 belongs to two other lanes; #6843 taken. Remaining engine-core work sits in untriaged findings (#7163 #7168 #7221 #7223 #7264 — five of them filed by this lane's own devs today), ⛔ which this seat cannot promote.

#6573 (PR #7210) MERGED as 708431313 — issue closed/completed, session archived, counted.
⚠️ Lesson ⑰ — removed_from_merge_queue is AMBIGUOUS. It fired on #7210 at 04:35Z and reads exactly like an ejection; the PR had in fact merged. A stale git log read a minute earlier said "not landed", which would have made it look like a kick. ⛔ Never call a kick from that event alone — read merged on the PR and git merge-base --is-ancestor on the commit. This seat nearly reported a false ejection.
#6546 (PR #7261) still in the queue, added_to_merge_queue 04:20Z, no removal.

#4797 DISPATCHED (session_015KREcaEs85dphzUUqdYfX6) the moment engine.ts freed up — held all shift on the file, never on the ruling. Brief carries the settled conditional-B ruling, ⛔ marked non-reopenable, while leaving genuinely open what the ruling does not fix: marker location, granularity, write path, and which gates count as irreversible. Also carries the ADR-0087 warning that cost #6573 a red (lesson ⑮).

#6017 veto window: no response yet on the ledger-only spec touch. ⚠️ Also found: os-help (services seat) filed the same-shaped declaration at 03:25Z for PR #7224 — a one-row error-code-ledger.zod.ts registration riding its own PR, with the same "registration is inseparable from the PR that mints it" reasoning. Two lanes independently converged on declare-and-proceed for ledger-only spec touches, which is stronger evidence for the convention than either declaration alone.

#6573 (PR #7210) — ACCEPT, ready + queued. Red cleared: dev took route 1 (kept major, filed the ADR-0087 disposition). 27 checks, 0 not-green. All four criteria verified by diff: both refusals (four spellings incl. the #5928 arithmetic one), ⛔ matching read unchanged (hookMatchesObject truthiness preserved; refusal at the registration door), caller survey self-measured, and the hook-exclude-objects.test.ts pin reworked (item 3 narrowed, new 3b) — content checked, not filename.
⚠️ New shape recorded — a LEDGER-ONLY packages/spec touch. #7210 writes migrations/registry.ts + spec-changes.json + the upgrade guide, and no schema surface at all. Measured: the last 8 commits touching that registry all also changed an authorable .zod.ts. No precedent either way ⇒ ⛔ did not rule on it myself; filed a cross-seat declaration on the domain:spec seat #6017 with an open veto window (#6532/#5586 precedent) and proceeded, as that precedent allows.

#6546 (PR #7261) — ACCEPT, queued. 34 checks, 0 not-green. Dev took the preferred route: typed the double IDataDriver instead of deleting the bit, restoring the tsc tombstone diagnostic that : any had switched off; supports: {}. Added the pin I asked for — "gates transactions on METHOD PRESENCE, not a capability bit" — running the transactional path against a driver advertising no capabilities. Changeset decision read from check-empty-changeset.mjs (skip-changeset label, ⛔ not an empty changeset — #4898 silent-publish trap). Swept and filed #7264 (~64 any-annotated doubles, same diagnostic off, none red today) instead of folding it in — card existence confirmed.

Next: ⛔ #4797 stays held until #7210 actually merges — not for the ruling (conditional B, settled 2026-08-06) but for the file: it needs objectql/engine.ts, which #7210 still owns while queued. ⚠️ #7210 touches ledger files that are a known ... is stale kick risk, so a kick would put the dev back in engine.ts; dispatching now would race it. Maintainer widened the mandate (04:0xZ): 「除了 v17,只要你车道的任务都可以处理」 — any in-lane card, not only target:v17.

In flight: #6573 (PR #7210, objectql/engine.ts) · #6546 (session_013oQBbBKaypak259s41pF2A, objectql/protocol-batch-atomic.test.ts).

Lane census 04:0xZ — 19 open issues: pm:queue 5 (#4707 #4797 #5299 #6546 #6752) · finding 7 · pm:on-hold 3 · pm:dispatched 2 (other seats) · other 2. Four of the seven findings were filed by this lane's own devs today (#7163 #7168 #7221 #7223) — all awaiting triage, ⛔ not this seat's to promote.

#4797 is READY and QUEUED BEHIND #6573 — not blocked on a ruling. It carries a maintainer-approved conditional B (2026-08-06) and a triage designation (not a split) naming this seat sole claimant: a lax-switch write records a deviation marker, and that marker must gate irreversible paths (the ADR-0104 field-file reclamation gate, which today reads only verified_at); reversible behaviour continues; a real os migrate --apply clears it. ⛔ Unconditional B (marker with no consumer) was explicitly rejected as "C plus a dead field"; ⛔ A rejected (makes the escape hatch one-way). Size M, pin tests required.
Why it is not dispatched yet: OS_ALLOW_LAX_* and the adr-0104-file-references gate both live in packages/objectql/src/engine.ts — the file #6573 is editing. Measured, not assumed (3 hits). The disjoint-surface rule caught it; it goes out the moment #6573 lands.

⚠️ #5299 needs care before dispatch: it spans driver-memory (frozen under #5499) and formula. ⚠️ #4707 spans three packages — dispatch when concurrency is low. ⛔ #6752 still needs a ruling.

#6725 (PR #7211) and #6678 (PR #7203) both merged 03:30:57Z; sessions archived; issues closed/completed.
#6573 (PR #7210) still open — one red: Check Changeset fails "Require an ADR-0087 disposition on a declared-breaking changeset" (the changeset declares major). ESLint + TypeCheck green. Dev poked with the diagnosis (its status wrongly said the PR had merged). ⛔ Two routes given, neither pre-ruled; ⛔ told not to weaken a refusal or downgrade the bump merely to pass a gate.

⚠️ Seat gap, recorded: the #6573 brief asked for a changeset on a public-API refusal and never named ADR-0087. Same shape as the gate-enumeration gap — confident about what it names, silent about what it omits. Lesson ⑮: a brief that requires a changeset for a breaking change must name the ADR-0087 disposition requirement.

🔵 In flight (surfaces disjoint by construction)

Card Session Surface
#6573 session_011rfAjYFr66Mjnz89gD5CL5 objectql/engine.ts — registerHook
#6725 session_0141cZum72My2vskaQSoQ1tZ objectql/metadata-facade.ts
#6678 session_01KbH3tZeQe1R4U8Z64eAF1m formula/parse-cel-to-ast.test.ts

✅ Wave 4 landed — three for three

Card PR Merged
#7049 target:v17 #7153 02:05:24Z bf42e76aa
#7047 #7161 02:13:39Z edb4af099
#7098 #7166 02:19:49Z 5d24f4b94

All three verified live on main, all three issues closed/completed, all three sessions archived. #7049 was this lane's first target:v17 card of the shift and it arrived through triage's routing channel — the case for refusing to cherry-pick other lanes' boards, made by outcome.

#7073 was NOT taken by this seat — it is now pm:dispatched to os-help. It had been held pending #7098 (shared file); another seat claimed it first. ⛔ Do not claim it.

v17 supply after wave 4: 0 (target:v17 ∩ domain:engine-core, counterprobes engine-core 22 / v17 8). #6843's ride-along trigger did not fire — #6725 touches objectql/metadata-facade.test.ts, not metadata/metadata-service.test.ts; it stays held and becomes standalone-eligible once this queue drains.

✅ Wave 3 closed — #6966 landed

#6966 / PR #7101 — MERGED 20:28:20Z as fc3a36af3. Verified live on main: hook.zod.ts:486 (dispatch), :795 (HookDispatch), both consumers (rule-hooks.ts ×3, file-reference-lifecycle.ts ×4), the export-origins entry, branch deleted. Both sessions archived (session_01EcLMsZRoR3daV3QzPXgwxK original dev — died at 16:20Z; session_01JPn1LbHwrvKXC1twtS6XvF requeue fix).

Two non-standard things happened and both are precedent:

batch:2 is safe here only because the two surfaces are disjoint — that was the selection criterion, not a coincidence. If either card grows into the other's files, it stops and reports rather than racing.

Pre-dispatch re-reads changed both cards' scope (full reasoning in each card's claim comment):

🎯 v17 — measured, not assumed

Maintainer asked to prioritise v17 three times (03:2xZ, 14:0xZ, 14:3xZ). Measured three times: label:target:v17 label:domain:engine-core = 0.

Repo-wide there are 12 open target:v17 cards, all in other lanes: domain:cli ×3 (#7033 #7023 #6599), repo:objectui ×4 (#6955 #6275 #5175 #5149), repo:cloud ×2 (#6954 #5852), domain:devx (#7005), domain:metadata (#6190), domain:spec (#4914). Most already assigned or queued by their own seats.

Standing ruling (maintainer 03:2xZ) remains in force: stay in this lane — ⛔ do not take over seat #6025, ⛔ do not cherry-pick v17 cards from other lanes. v17 work enters this lane only via triage labelling. #6832 is the proof the channel works: filed here, graded target:v17 by triage at 03:29Z, shipped 07:38Z.

Referred to triage for grading without lobbying: #6966 (drift introduced by #5574, which shipped under target:v17 in PR #6697) — flagged honestly as materially weaker than #6832, since the card's own measurement shows no user-visible failure today.

⏭️ Next up

#6573 / #6546 / #6725 / #4707. ⛔ #6752 needs a ruling before it can be dispatched ("需定夺是否一并退休"), so it is not queue-ready.
#6843 — hold with a falsifiable trigger: rides along with any card touching packages/metadata/src/metadata-service.test.ts (disposition 1 pre-approved, 2/3 rejected); if nothing touches it before this lane's queue drains, promote it standalone as XS.

🏆 Shift tally (since 08:46Z)

19 MERGED: #6658 (#5896, predecessor), #6670 (#4840), #6672 (#5282), #6677 (#6223), #6680 (#5586), #6671 (#1825), #6697 (#5574+#5846, XL, target:v17), #6766 (#6132), #6786 (#5543, target:v17), #6788 (#6457, target:v17), #6791 (#4776), #6794 (#5929), #6812 (#6810), #6818 (#6808), #6454 (confirmation), #6907 (#6806, target:v17 inherited from #5495), #6972 (#6832, target:v17), #7097 (#6679), #7101 (#6966).

#7097 / #6679 acceptance record (16:05:16Z, merge commit 29b94ed2a; session archived): 25 check-runs, 0 not green; 3 files, all packages/formula; the UNSOUND_OVERLOAD_RE scope guard held. This card's grading was wrong and the implementation proved it — the filing marked the impact "appears nil" while explicitly flagging one direction as unmeasured; triage graded it not-target:v17 on that basis and this seat forwarded that framing into the dispatch brief. Measuring it for the fix found the case: when hydration lets the expression short-circuit around the throwing call, the spurious retry succeeds and returns a value where the fault was the right answer. Referred back to triage for target:v17 re-grading.

#6972 / #6832 acceptance record (07:38:34Z, merge commit 55011afa3): all eight step-7 criteria passed; grounds in the ACCEPT comment on #6832. The dev chose shape (b) "derive from schema" and excluded (a) by measurement — HookSchema.safeParse refuses four shapes the binder binds today (unknown key, unknown event name, wrong type, alias retries), so (a) would turn "binds and runs" into "fails to bind": blast radius larger than the defect. It also pinned "explicit 0 still wins" (the ??/|| trap) unprompted, and resolved the schema lazily because lazySchema exists precisely to avoid that allocation.

Handover still open: #6916 (domain:drivers) — driver-memory collisions can only ever land as silent duplicates; the only real fix sits inside the #5499 freeze; the engine-side alternative was measured and rejected, noted so nobody re-proposes it. #6916 is also the third piece of the #5495 closeout question.

📋 Other ledger

Operating notes (still in force)

curl -s -H "Authorization: Bearer $GITHUB_TOKEN" \
  'https://api.github.com/repos/objectstack-ai/objectstack/issues?state=open&labels=target:v17,domain:engine-core&per_page=100'

Measured 22:5xZ with both counterprobes green (domain:engine-core alone 21, target:v17 alone 9): the intersection is 0 — this lane's v17 supply, fourth measurement. (The MCP search_issues tool does work; only the raw REST search path is blocked.)

git show origin/main:.github/workflows/lint.yml | grep -oE 'check:[a-z0-9:-]+' | sort -u   # 64

64 unique check:* in lint.yml, 70 across all workflow files ⑤ four disciplines for parallel work in one file (three cards, zero conflicts) — extended in wave 3 to pair cards by disjoint surface when running batch:2 ⑥ label PM assumptions "must be measured, falsification welcome" (three times, three payoffs — #6832's was the strongest) ⑦ archive the cloud card the moment its PR merges ⑧ a split card inherits its parent's release label (#6806 / #5495) ⑨ close out the wave, then compact ⑩ when a filer marks an "impact nil" grade as unmeasured in one direction, the dispatch brief must carry it as an OPEN QUESTION, never repeat it as a finding — #6679: the dev measured a wrong value while under a brief telling it not to inflate the card, which is what a correct response to that instruction looks like. ⑪ a site the card calls "comments only, no behaviour" is briefed as TO BE VERIFIED, not as settled — #6966: site 3 was in fact a user-visible failure on an access-control surface (every bulk write to a ruled object revoking all its rule grants), and the dev is who falsified it. ⑫ a live-enumeration rule is only as good as its command — state the command, and re-check it against the file it reads; a systematically blind grep is worse than no rule, because it yields a confident wrong number everyone downstream trusts (see ④).

Generated by Claude Code

Metadata

Metadata

Assignees

Labels

pm:seatPM seat registry issue - single-writer body, index = this label

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions