Hypothesis: If the model’s input features look materially different for grade A/B deals that went UTP versus those that did not, it may suggest the model is not weighting those features appropriately. Conversely, if the feature distributions are similar, the model may lack the information needed to separate these cases.
Written in simple english..