A principle in software is to never write the most complicated code you possibly can, because maintenance is harder than writing, and if you could only just barely write it in the first place, you won't be able to maintain it.

The principle for pilots is similar.

As to skills assessments, you're right, it's difficult. But you can use some general proxies. Certification pilots should be drawn from a standard pool, not a test-pilot pool. Some sense of performance on some standardised battery of qualification flights or simulator check-flights might be used, with scoring based on success or performance in identifying and addressing conditions encountered.

The fact that a precise quantity cannot be confidently stated doesn't mean that no quantity or range can be given.

And for the sake of the certification pilots, I'm assuming check-out flights would either be in simulators or with a more qualified check pilot assisting.