Usually you draw a representative random sample for these kinds of things. You only test every bar if you detect a problem in the random sample.