Race conditions in places like this are exceedingly hard to write reliable tests for. It may take one, two boots; dozens, or thousands, or you may just get insanely lucky and whatever arbitrary # boots you do to try and reproduce it was still simply not enough. It's hard to have any level of confidence, in many cases.

There are acceptance tests, intergration test and actually many other types. I'm similarly not familiar with kernel dev. Have studies both operating systems and distributed systems aswell as CS. But the build checked all the right boxes.