I don't remember much about model algorithms, but it seems that when they change dramatically between 6 hr runs, there has to be some reason. I guess, 5-6 days out, the potential outcomes are so great that it can spit out 2 very different ones based on permutations alone.