At a certain point you just need to take a statistically representative sample of the output for review.