I think the content about decision boundaries and that 538 doesn’t state an outcome misses the point of stating probabilities. The better way to judge the forecasts is using something like a Brier score. Silver’s 90% predictions should match reality 90% of the time, his 50% predictions should match reality 50% of the time, etc. This includes all of the predictions from six months out until election day itself. I think Silver would agree that he should not be judged solely on the election day probability as you suggest.