[CI][Windows] ArtefactExtractionServiceTests.ExtractAsync_ShouldPropagateCallerCancellationWithoutRecording fails intermittently on Backend Unit (windows-latest)

[Chris0Jeky]

What happened

Required CI job Backend Unit / Backend Unit (windows-latest) failed on PR #2627 (a one-file frontend change, PaperToastContainer.vue, no backend file touched) at head 28890ff63:

• run 33941869440, job 101240765582, 2026-09-05 03:28Z
• Taskdeck.Application.Tests.Services.ArtefactExtractionServiceTests.ExtractAsync_ShouldPropagateCallerCancellationWithoutRecording [FAIL] after 5 s

One diagnostic rerun of the failed job (same head, 13:2xZ) passed, and the PR merged as 6d34e36b4. The diff cannot have caused it, so this is a timing case in the caller-cancellation propagation test on the Windows runner.

Why it matters

Every occurrence costs a rerun on an unrelated PR and, under ci-required.yml's concurrency group, can hide behind a cancelled main run. There was no tracked issue for this shape (searched ArtefactExtractionServiceTests and PropagateCallerCancellation across open and closed issues on 2026-09-05), so occurrences were not being counted.

Acceptance

• Read the test and the service path it exercises; say whether the 5 s figure is a test-side timeout (WaitAsync, Task.Delay) or a runner stall, and whether the cancellation is observed through a token that the test controls.
• Make the test deterministic (signal-based rather than time-based) or, if the runner is the cause, classify it under CI-15 (CI-15: Fail-visible flake classification, quarantine with expiry, and retry governance #2339) flake governance with the log evidence above.
• Record further occurrences here (run id, job id, PR) until the fix lands.

Refs #2339, #2378 (the sibling Windows timing class on the Frontend Unit leg).

[Chris0Jeky]

Coordinator disposition (2026-09-06, Codex-session PR carried under the standing D-12 route: fresh-context review, MEDIUM+ fixed in-PR, merged as a merge commit).

PR #2716 (test: synchronize extraction cancellation) merged as 8a6f64211; review SHIP, MEDIUM-1 fixed in-PR (the assertion await is bounded to 30 s again so a propagation regression fails red instead of hanging the job; class re-run locally 18/18). Kept open for observation: the 5 s worker-entry wait this issue tripped on is retained (async now, no pool thread held), so the fix is a plausibility argument until the Windows Backend Unit leg stays green across the next runs. LOWs: dead signalStarted stub parameter; docs/releases/V0_3_0_READINESS.md:128 names the deleted test.