How do I catch failures that don't reproduce when validating a design?

You can validate most designs. What defeats you are the failures that appear once and then vanish when you go looking. A bug you cannot reproduce feels like chasing a ghost, and it is genuinely one of the hardest things in the work. Struggling with it is not a shortfall in your skill. A failure that will not reproduce is not random, it only looks that way because you cannot yet see the condition behind it. Something differed that run, timing, a specific order, a value you were not watching, a state left over from before. The work is to capture more of the scene when it happens, logs, inputs, the environment, so the next occurrence hands you the clue this one hid. Make the ghost leave footprints. Next time one appears, resist the urge to move on because it went away. Spend the time to add logging around where it struck, so if it returns you will catch it in the act. You solve these by patiently narrowing what could differ, not by luck. One captured case cracks the whole class.