Many bioinformatics applications utilize machine learning techniques to create models for predicting which parts of proteins will bind to targets. Understanding the results of these protein surface binding classifiers is challenging, as the individual answers are embedded spatially on the surface of the molecules, yet the performance needs to be understood over an entire corpus of molecules. In this project, we introduce a multi-scale approach for assessing the performance of these structural classifiers, providing coordinated views for both corpus level overviews as well as spatially-embedded results on the three-dimensional structures of proteins.