Comment by jasonmarks_

More objectively it reads as if none of the models performed well outside of 24 hours with a significant uptick in inaccuracy after 72 hours.