The problem is using statistical models at all for predicting the probability of one time events. (elections are every 4 years, but those future models will be

The problem is using statistical models at all for predicting the probability of one time events. (elections are every 4 years, but those future models will be different). You can't validate such a model, and it's not useful since any upset victory was "accounted for" simply because its possibility was named.