check-engine-double-contract still cannot read 21 constructs at the local-binding and shorthand spellings — measured at 7 files needing a pin, 0 needing a baseline raise

[os-zhuang]

Filed unassigned by the dev seat working #9877, whose PR takes the first of the
three spellings that card names (the ?? / || default) and deliberately narrows away
from the other two. This is the remainder, with the price measured rather than estimated.

What is left

After #9877's PR, UNRECOGNISED stands at 21, in two spellings — both of which need
localFunctions-style resolution rather than a one-line unwrap:

spelling	count	example
the initializer roots at a binding the file declares	7	delete: del, with const del = vi.fn(async () => ({ deleted: 1 })) above it
shorthand member	14	const engine: any = { registry, insert, findOne, update, delete: del }

The price, measured on claude/issue-9877-engine-double-recogniser

Measured with a throwaway probe that resolved an Identifier initializer and a
ShorthandPropertyAssignment through the file's own bindings, then reverted. Numbers
from the gate's own output, not from a hand count:

• +10 constructs enter the discovered population (delete 232 → 238, update 265 → 269).
• 7 PINNED errors, in 7 files, all pinnable at source:
  • packages/metadata-protocol/src/protocol.batch-atomic.test.ts [delete]
  • packages/metadata-protocol/src/protocol.batch-not-attempted.test.ts [delete]
  • packages/metadata-protocol/src/protocol.bulk-record-not-found.test.ts [delete]
  • packages/metadata-protocol/src/protocol.many-data-atomic.test.ts [delete]
  • packages/metadata-protocol/src/protocol.record-not-found.test.ts [delete]
  • packages/metadata-protocol/src/protocol.dropped-fields.bulk.test.ts [update]
  • packages/plugins/plugin-auth/src/admin-user-endpoints.test.ts [update]
• 2 further RETAINED rows on protocol.update-path-id-wins.test.ts — new pinned
  coverage the ledger has to learn, i.e. a --write, not a problem.
• ⭐ 0 of them require raising the shrink-only baseline. That is the headline: the
  guardrail that made check-engine-double-contract cannot read three mock-initializer spellings, and one of them hides a live unguarded engine delete double #9877 expensive does not bind here either. This is pinning work in
  two packages, not a ratchet negotiation.

Why it was not folded into #9877's PR

Ruling on that dispatch was "pin the newly-visible doubles, or narrow the widening to
what you can pin". The ?? half is two constructs in one file and was pinned outright;
this half is seven test files across two packages, each needing the dispatch predicate
wired into a fake whose call shapes those suites drive deliberately (not-attempted,
record-not-found, bulk) — a different act with its own verification surface.

One structural note for whoever takes it

⛔ Do not do what the probe did. It patched scanSource's consider only, which left
censusSource still reporting the same constructs as UNRECOGNISED — discovery and the
census then disagree, and #9747's self-test asserts they agree. The real change moves
implOf itself, which means threading the SourceFile (or a resolved binding map) into
it and through both walks. That is the reason this is not a one-line follow-up.

node scripts/check-engine-double-contract.mjs --census (added by #9943's commit on the
same PR) prints the per-kind read rates, so the 34 identifier and 21 shorthand members
are re-derivable rather than quoted from this card.

Related: #9877 (parent) · #8639 (the same blind half at vi.fn(fn)) · #9747 (the census
that made this countable)

---

Generated by Claude Code

[os-zhuang]

Triage: lands in scripts/check-engine-double-contract.mjs (gate tooling) ⇒ domain:devx, type Task. State: pm:blocked —

Blocked-by: #9877

Rationale: the measured premise (UNRECOGNISED at 21, the --census flag, the 7-pin price) is taken on #9877's PR branch, and the structural note (implOf threading) builds directly on that PR's recogniser; same-file hard serial applies regardless. Unlocks when #9877's PR lands; re-verify the census numbers on the merged ref before dispatch.

---

Generated by Claude Code

[os-zhuang]

Unlock scan (triage seat): pm:blocked → pm:queue. The block's stated condition — "#9877's PR lands" — has fired: PR #10176's content is on origin/main at 359f5956d (verified by content, --census present in scripts/check-engine-double-contract.mjs, 27 hits; ⛔ not by shallow-clone git log --grep, which is blind here). Release double-check passed: this is the most recent (only) conversion comment's condition, and the card carries no newer merged PR.

Carried into dispatch per the conversion comment and the card's own structural note: re-derive the census numbers (node scripts/check-engine-double-contract.mjs --census) on the merged ref before acting — the 21/7-files/0-baseline figures were measured on the PR branch; and the fix must move implOf itself (threading the SourceFile/binding map through both walks), not patch scanSource's consider alone, or discovery and census disagree and #9747's self-test reds. The #9877-dispatch guardrail is inherited verbatim: ⛔ never raise the shrink-only baseline; pin or narrow, and report needs_decision with the enumeration if neither is honest.

Size/model suggestion: M, mode:subagent, opus.

---

Generated by Claude Code