[engine-double-contract] Pin the 29 baseline entries #8639 first-measured — each needs a signature widening, a tripwire judgement, a body conversion, or one devDependency

[qq9340100]

Follow-up to #8639 (PR #8650). Filed unassigned; nobody is on it.

Blocked-by: #8639

#8639 taught implOf to unwrap a mock constructor's sole function argument, which moved
check-engine-double-contract's discovery from 379 doubles to 435. Of the 56 newly discovered
engine-shaped doubles, 21 were pinned in that PR — every one whose pin was a single call —
and 29 (file, verb) pairs entered scripts/engine-double-contract.baseline.json as first
measurements. This card is the remainder.

Same shape as #5855, which closed the six entries #5619's sinking had unlocked.

Why each was not pinned in the same pass, measured

The 29 pairs fall into four classes, and the class is what decides the work:

• Signature widening (the bulk). The implementation declares fewer than two parameters, so it
  has no binding to hand the predicate — delete: vi.fn(async () => ({ deleted: 1 })). A pin
  must first add the options bag / payload the fake omits, which changes the double rather than
  routing it. Mostly packages/metadata-protocol/src/seed-loader-*.test.ts.
• Deliberate tripwire doubles. protocol.validate-data.test.ts and two doubles in
  seed-loader-driver-text.test.ts open their body by throwing, because the test asserts the path
  performs no write at all. Routing them through the predicate changes which error the assertion
  sees, so pinning needs the author's intent read, not a mechanical insert.
• Concise arrow bodies. protocol.dropped-fields.test.ts and seed-loader-env-scope.test.ts
  have expression-bodied arrows — no statement list to insert in front of, so the body must be
  converted to a block first.
• @objectstack/mcp (6 pairs). The package depends on neither @objectstack/metadata-core
  nor @objectstack/objectql, so a pin is not one call: it needs a devDependency first. Those
  implementations also declare no parameters, so they need widening too.

Notes for whoever takes it

• Each baseline entry already carries its own why and closes naming its specific obstacle.
• These entries are first measurements, not a raised ratchet — the doubles were always there,
  the scan could not reach them. Removing an entry is the ratchet tightening, squarely the
  author's job; the baseline is shrink-only and the checker reconciles in both directions, so a
  pinned double's entry must be deleted in the same PR.
• Worth doing per package rather than in one sweep: each pin can flip a suite red, and a red here
  is a real finding (exercised looseness), not a nuisance to be worked around.
• All four packages touched by the fix(scripts): check-engine-double-contract discovers doubles whose write verb is vi.fn(...) #8650 pins were run green there, so the harnesses are known good.

[hotlong]

Triage: pm:queue + pm:blocked (Blocked-by: #8639 added to the body) + domain:devx, type Task — cross-domain exception card, devx executes.

Rationale:

• Blocked, verified: the 29 baseline entries and the widened discovery exist only on PR fix(scripts): check-engine-double-contract discovers doubles whose write verb is vi.fn(...) #8650's branch (draft, not merged, checked at triage time). The card is implementable the moment fix(scripts): check-engine-double-contract discovers doubles whose write verb is vi.fn(...) #8650 lands; the unlock scan will requeue it when check-engine-double-contract never discovers a double whose write verb is vi.fn(…) — implOf drops a call-expression initializer #8639 closes.
• Route: the landing files span three lanes' packages — packages/metadata-protocol test files (bulk: seed-loader-*.test.ts, protocol.validate-data.test.ts, protocol.dropped-fields.test.ts), packages/mcp (6 pairs + one devDependency + package.json), and the shared shrink-only ledger scripts/engine-double-contract.baseline.json, which every pin PR must edit (entry deletion in the same PR). Splitting per lane would put multiple seats on that one ledger file; keeping one executor with per-package serial PRs (as the card itself prescribes) is the safer shape. devx is designated because the invariant being restored is the devx-owned gate's ledger discipline and the check-engine-double-contract never discovers a double whose write verb is vi.fn(…) — implOf drops a call-expression initializer #8639/fix(scripts): check-engine-double-contract discovers doubles whose write verb is vi.fn(...) #8650 context lives in that lane. Expected file surface for the claim comment: the named test files above + scripts/engine-double-contract.baseline.json + packages/mcp/package.json (+ lockfile).
• No human floor triggered: the needed devDependency is workspace-internal (@objectstack/metadata-core or @objectstack/objectql), not a new third-party runtime dependency.
• Dispatch notes: the tripwire-double class is an intent-read judgement, not mechanical — dev must not convert a deliberately-throwing double in a way that changes which error the assertion sees; a suite flipping red on a pin is a real finding to report, not to suppress. Baseline entries' own why/closes fields are the per-entry spec.
• Dedup clean: [finding] check:engine-double-contract is a PM judgement call in dispatch-gates.mjs, not a convention entry — and a mechanical trigger for it exists #8632 (skills lane) concerns dispatch-gates wiring of this checker, not these entries; [engine-double-contract] #5619 的下沉解锁了另外 6 条被同一个环卡住的条目(core / metadata / platform-objects)—— 现在各是一行 pin #5855 is the closed precedent shape.

本评论来自分诊座位 Routine,不构成认领。

---

Generated by Claude Code

[claude]

Half-state repair: removed pm:queue, kept pm:blocked.

This card carried pm:queue and pm:blocked at once — two pm-states on one card, which makes it both selectable and un-selectable depending on which query runs. Found during the 2026-08-14 unlock sweep over every pm:blocked card in the three repos.

Resolved toward pm:blocked, because the block is real: the declared blocker #8639 is open. This card was not part of the 23 unlocked in that sweep, and dispatching it now would hand a dev work whose upstream has not landed.

When #8639 closes, this becomes a one-label flip to pm:queue.

Authorization: maintainer directive (this session) —「先做 23 张 blocked 解锁 10 张 on-hold 按各自自述处理」· PM session 01JaVVMrSxt7Tgi1uwEuDtH7

---

Generated by Claude Code

[hotlong]

Unblocked — pm:blocked → pm:queue. Triage seat Routine, unlock scan of 2026-08-14T16:5xZ.

Upstream #8639 closed completed 2026-08-14T12:23:58Z via merged PR #8650.

Premise re-verified on the merged ref, not assumed. The unlock rule is that the merge which closes the upstream is the merge most likely to have absorbed the downstream too, so the card's file surface was re-measured against origin/main rather than trusted:

git show origin/main:scripts/engine-double-contract.baseline.json
→ 164 entries total
→ entries citing #8639: 29 (all kind=DEBT)
→ by package: metadata-protocol 19 · mcp 6 · plugins 2 · objectql 1 · runtime 1

29 — exactly the count this card claims, all still DEBT, none pinned by #8650 itself. The residual is intact and the card stands unchanged. The package split also matches the body's four-class analysis (the mcp 6 are the ones needing a devDependency first).

No scope change; no re-routing. domain:devx retained.

---

Generated by Claude Code