Context
The Approve+Follow-Up obligation from PR #16504's review (Emmy, 2026-08-04 — her Depth Floor falsified the shipped causal story while approving the seam; the operator's SharedWorker concern was the falsifier she chased). Filed at merge per the A+FU contract. Live latest-open sweep at filing: latest 20 checked, no equivalent.
The Problem
PR #16504 shipped the RIGHT repair (resolve Neo.currentWorker at call time in both theme-deferral gates) wearing the WRONG law. The source comments, ticket #16503, and PR body claim: "test runners reuse worker processes across spec files and reassign the world between them" / "a later spec file's setup() builds a fresh app world and reassigns Neo.currentWorker." Both claims are falsified by the exact tree:
test/playwright/setup.mjs assigns via Neo.currentWorker ??= — it never replaces an existing worker.- The true mechanism is module evaluation before harness setup:
BodyContainer.spec.mjs statically imports the component/VdomLifecycle graph before its top-level setup() executes, so the module-load binding captured undefined; the old gate then threw TypeError: Cannot read properties of undefined (reading 'countLoadingThemeFiles'), and Overflow.syncControl()'s catch swallowed the rejected mount — presenting as "0 listeners" and hiding the import-time undefined.
- The production framing is also wrong in the comments: a surviving SharedWorker realm keeps the SAME App-worker object on reconnect (WHATWG matching/lifetime); realm death refreshes modules AND singleton together. There is no production world-reassignment the comment's "law" warns about — the issue is captured-identity-possibly-undefined at import time, never "stale mutable state of a previous world" (a captured object reference always exposes live mutable fields).
The Architectural Reality
src/mixin/VdomLifecycle.mjs: the binding-site guard comment and both gate comments carry the false story; the gates each read Neo.currentWorker twice (count consult + listener registration).- Reviewer's non-blocking shape suggestion, adopted here: per gate, one method-local
const worker = Neo.currentWorker, then worker.countLoadingThemeFiles / worker.on(...) — one coherent owner, one global read.
#16503's close-out comment carries the corrected mechanism; this ticket corrects the SOURCE surfaces.
The Fix
- Rewrite the three comment blocks in
VdomLifecycle.mjs to state the true mechanism (import-time capture may precede harness setup ⟹ possibly undefined; call-time resolution is the contract) and drop the false worker-reassignment/previous-world-state claims.
- Apply the single-read shape at both gates (
const worker = Neo.currentWorker).
- Optional hardening witness if cheap: a unit probe pinning the cold-import-order case (module graph evaluated before
setup()), so the swallowed-TypeError presentation can never hide this class again.
Acceptance Criteria
Out of Scope
- Any behavioral change to the theme-deferral contract (proven by the
#16434 five-beats E2E).
- Harness import-order policy.
Decision Record impact
none.
Related
#16503 / PR #16504 (the shipped seam + the review carrying the falsification), #16434 (the contract's origin), PR #16489 (first CI casualty).
Origin Session ID: 1913de09-6dc0-4d1e-a9a3-b51c33b46cdc
Retrieval Hint: query_raw_memories("theme deferral causal story correction import-time undefined module evaluation before setup single worker read")
Context
The Approve+Follow-Up obligation from PR
#16504's review (Emmy, 2026-08-04 — her Depth Floor falsified the shipped causal story while approving the seam; the operator's SharedWorker concern was the falsifier she chased). Filed at merge per the A+FU contract. Live latest-open sweep at filing: latest 20 checked, no equivalent.The Problem
PR
#16504shipped the RIGHT repair (resolveNeo.currentWorkerat call time in both theme-deferral gates) wearing the WRONG law. The source comments, ticket#16503, and PR body claim: "test runners reuse worker processes across spec files and reassign the world between them" / "a later spec file's setup() builds a fresh app world and reassigns Neo.currentWorker." Both claims are falsified by the exact tree:test/playwright/setup.mjsassigns viaNeo.currentWorker ??=— it never replaces an existing worker.BodyContainer.spec.mjsstatically imports the component/VdomLifecyclegraph before its top-levelsetup()executes, so the module-load binding capturedundefined; the old gate then threwTypeError: Cannot read properties of undefined (reading 'countLoadingThemeFiles'), andOverflow.syncControl()'s catch swallowed the rejected mount — presenting as "0 listeners" and hiding the import-timeundefined.The Architectural Reality
src/mixin/VdomLifecycle.mjs: the binding-site guard comment and both gate comments carry the false story; the gates each readNeo.currentWorkertwice (count consult + listener registration).const worker = Neo.currentWorker, thenworker.countLoadingThemeFiles/worker.on(...)— one coherent owner, one global read.#16503's close-out comment carries the corrected mechanism; this ticket corrects the SOURCE surfaces.The Fix
VdomLifecycle.mjsto state the true mechanism (import-time capture may precede harness setup ⟹ possiblyundefined; call-time resolution is the contract) and drop the false worker-reassignment/previous-world-state claims.const worker = Neo.currentWorker).setup()), so the swallowed-TypeError presentation can never hide this class again.Acceptance Criteria
VdomLifecycle.mjsclaims worker-world reassignment or previous-world mutable state; the stated mechanism matches the review's falsified-and-corrected account.Neo.currentWorkerexactly once per invocation via a method-local.31 passed/26 passed); behavior byte-identical.#16504gate shape.Out of Scope
#16434five-beats E2E).Decision Record impact
none.
Related
#16503/ PR#16504(the shipped seam + the review carrying the falsification),#16434(the contract's origin), PR#16489(first CI casualty).Origin Session ID: 1913de09-6dc0-4d1e-a9a3-b51c33b46cdc
Retrieval Hint:
query_raw_memories("theme deferral causal story correction import-time undefined module evaluation before setup single worker read")