To what extent are the models in questioned developed using historical data? The models could get good results on hindcasting because they make good predictions, or they could get good results on hindcasting because they have been tuned for that data-set already, meaning that the good hindcasting doesn't tell us much. Is it possible to distinguish these?