Comment by andai

Results seem mostly noise to me. One eval per model, in a large problem space (i.e. a problem which requires many attempts to solve well).