Morehead State's beating Louisville points up a limitation on all predictive models, whether mathematical or observational. At best, even exceptional predictions can only identify which team would win the majority of times if they played multiple games against each other. If Louisville plays Morehead State five times, it wins that matchup 3 or even 4 times. And that is all a method can ever do for you. It can't tell you whether this is going to be the one time out of five that the underdog puts it all together and wins.
How often does the better team win in college basketball? 75-80% of the time? That's the best you will ever get from a predictive model. Anything beyond that has a lot of luck involved. Not blind luck, of course. It is meaningful to identify the teams that have a better or worse chance than the common wisdom. But even if you identify a 12 seed that really is a better team and should beat the 5 seed 60% of the time, they are still going to lose 40% of the time.