The problem of scoring protein models for some fit criteria, such as closeness to native structure or ability to bind, is one of the earliest problems recognized in the field of proteomics. Here, an approach capturing properties of the entire protein and utilizing this as input features for a machine learner will be presented. This approach allows us to better match the conditions of the Boltzmann approach that connects phase space states probability and energy. We test our approach on some recent results from CASP experiments. A partial review of this presentation is available from: Faraggi, Eshel, and Andrzej Kloczkowski. Proteins: Structure, Function, and Bioinformatics 82.5 (2014): 752-759.