Our offline handler accepts an old completion against whichever fixture-check attempt is current. Random sleeps reproduce it only sometimes. I want to choose the event order deliberately and check both attempts afterwards.
Can the test queue the old completion explicitly, start the new attempt, and then deliver the queued event? That would let you assert the old attempt's state and that the new one is still waiting, without racing the clock.
We added an explicit event queue. The sequence fails every time now: start first attempt, retain its completion, start second, deliver the first completion. New attempt wrongly becomes complete.
Keep that reproducible failure in the handover even before the fix. Add a duplicated delivery and a genuine completion for the second attempt as separate cases; the person maintaining it needs to see what should remain unchanged and what should actually advance.