*I* use a representative sample for the final approval of an in-line or final inspection device. A representative sample means that I have the same percentage of good and bad units as in production. I try to get the variation in goodness and badness to also be proportionally representative if and with the same mix of defect types when appropriate. Thsi can make the sample size rather large but it provides the most information of how many actual false positives and negatives you will get.
If I don't knwo this yet or the samples / testing are really hard, I will reduce the sample size and concentrate on the marginally good and marginally bad.