The four tests that retry a toggle after a failed first try advanced the fake clock by RETRY_FIRST, but the wait before a second try is RETRY_FIRST doubled once (2000 ms). The retry only fired because shouldAdvanceTime let real time push the clock the rest of the way inside waitFor's 1000 ms timeout, a race that CI lost once (run 3860). retrySecondTry now waits until that 2 * RETRY_FIRST timer is set, then advances past it, so the jump can no longer land before the timer. The no-online-event test also asserts nothing was sent before the timer fires. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>