Evaluating global models