Subpart error distributions as 2D histograms against relevant target quantities or scales. For each subpart, the corresponding 1D error histogram and the associated score response functions are provided. The dashed red lines indicate the error region where the score saturates at 1. Orange lines and arrows indicate the range of values from the training sample when relevant. Dotted black lines represent the error obtained when predicting the minimum value from the training sample across the full comparison scale.