Makes sense. It might be useful to score models based on how much they improve your predictions, e.g. Newtonian physics is not completely accurate but improves my predictions substantially vs not knowing any physics.