Predicting probability based on mean and failed trials