To again be fair to 538, they are pretty clear that their model is based on polling data, at least in their "polls only" model, which I think they put more emphasis on than their "polls plus" model. Maybe next cycle, they'll do more work on that plus model to account for more factors or maybe this truly was an outlier election, which happens every so often.

This is the fundamental problem with assigning probabilities to events that will only happen once. It doesn't make any sense!

Sometimes there are no definitive answers. The world we live in is not deterministic, at least with our current knowledge of the world.