The GitHub-integration recovery loop, the delayed PR refresh, and the toast auto-dismiss slept on a real `ContinuousClock()`, so a TestStore could not control them. Under parallel CI the 15s recovery interval could elapse inside a test's window and emit an extra `refreshGithubIntegrationAvailability` -> `githubIntegrationAvailabilityUpdated` pair, failing `worktreeInfoEventRepositoryPullRequestRefreshQueuesWhileAvailabilityUnknown` with "Must handle 2 received actions before sending an action". It passed locally and in isolation, so it read as noise, but it is a real non-deterministic-clock defect. Inject `@Dependency(\.continuousClock)` and sleep on it at all three sites; the affected tests provide a `TestClock`, so the delays never fire unless advanced. Claude-Session: https://claude.ai/code/session_01AQ4X2bx8DnU74wwThhV9D8