As in any test methodology, if your random sample has problems, it's a bad sign for failure rates for the batch.