Evaluation is AUC (are the model's hit-probabilities for bona fide hits typically higher than the model's probabilities for flops?