Problem
packages/queue/tests/typed-queue_test.ts:31 is nondeterministic in the repo-wide suite. During PR #1665 verification at immutable SDK-only source head 9a26c107a, one root run ended 4202 passed / 1 failed / 19 ignored: the DLQ assertion expected one item but observed two. The PR changes nothing under packages/queue.
The same test passed when run alone, passed when paired with the new SDK KV test, and later repo-wide runs at the identical head ended 4203 passed / 0 failed / 19 ignored. The formal evaluator also ran the root suite once at that same product head and did not hit the failure. This is a timing/state-isolation flake, not an SDK regression.
Impact
- creates false-red required CI and forces costly attribution/reruns;
- wastes coordinator and evaluator time and tokens;
- can hide a real queue regression behind an accepted "known flake" narrative if left unowned.
Acceptance
Evidence
Observed during NetScript 0.0.7 PR #1665 Tier-A/formal evaluation. The branch was SDK/docs-only; isolated and identical-head reruns distinguish this from the leaf delta.
Problem
packages/queue/tests/typed-queue_test.ts:31is nondeterministic in the repo-wide suite. During PR #1665 verification at immutable SDK-only source head9a26c107a, one root run ended4202 passed / 1 failed / 19 ignored: the DLQ assertion expected one item but observed two. The PR changes nothing underpackages/queue.The same test passed when run alone, passed when paired with the new SDK KV test, and later repo-wide runs at the identical head ended
4203 passed / 0 failed / 19 ignored. The formal evaluator also ran the root suite once at that same product head and did not hit the failure. This is a timing/state-isolation flake, not an SDK regression.Impact
Acceptance
Evidence
Observed during NetScript 0.0.7 PR #1665 Tier-A/formal evaluation. The branch was SDK/docs-only; isolated and identical-head reruns distinguish this from the leaf delta.