Our offline inspection handler receives an old result after a new attempt starts and marks the new one complete. Both attempts use the same job label. Random delays make the test unreliable. I want the exact event order under test control before anyone calls this fixed.
Queue the earlier result in the test and release it only after the new attempt exists. Assert both attempts' stored states, not just the current screen. Does the result carry an attempt identity that the handler is ignoring?
It does. The handler looks up the current attempt by job label instead. We now deliver the old result explicitly, and the test fails in the same way every run.
Then make the repair use that carried identity, including a genuine late result for an older attempt. Don't fix it by throwing away everything old; that earlier attempt may still need its result reconciled.