I'm not super happy with the pattern of thinking in these numbers; arguing 1/100000 vs 1/100 for "accidents" is again a boolean thing. They probably had a very specific definition, which does make this viable, but we don't have that definition here. So the numbers are meaningless to us. And not having that definition, "accidents" is a sliding scale… e.g. I'm pretty sure astronauts injured themselves banging various bodyparts against various parts of the spaceship. That's technically an accident.

And in this concrete example — the heat shield isn't boolean either. I don't know how steep the gradient between pass and fail is, but it certainly exists, and it's possible they come back successfully but with it singed significantly outside expected parameters. (Even "less than expected" would indicate a problem here IMHO.) That does mean it's not necessarily a question of having enough boolean datapoints.

A high enough threshold for "accident" is a death. In the space shuttle the "accident" rate was 2/135 that is somewhat close to 1/100, but it shows that 1/100000 is too optimistic.

You can pick a more strict safety criteria, but the result would be (something)/135 >= 2/135 > 1/100 >> 1/1000000