A Bayesian approach for on-line max and min auditing