I can make the offline inspection tracker complete the wrong attempt by delivering an old event after a new start, but random sleeps make the result awkward to repeat; how should I control event order and identity, including reuse after restart, without giving the test knowledge the real interface lacks?