When pressed before the election, Silver did not explain where Trump's much higher probability of winning came from. He predicted a Trump loss, Trump won, and he claimed victory because he gave Trump a better chance of winning. There's no way that strategy could have failed.
Silver claimed that his model was better because it predicted a high correlation between PA/MI/WI.
A model that predicts a 30% chance of winning the election will be wrong 1 out of 3 times, which is not quite a coin flip but close enough.
Nate Silver is not a magician! He can't magically make polls reliable!
All he (or anyone) can do is interpret or analyse poll results, and then surface their findings in a way a larger audience can understand. 538 did that better than any other poll analyst ... but they all got it wrong because the polls themselves were faulty.
TLDR; You can't get water from a stone, and no one (not even Nate Silver) can get perfectly accurate predictions from (inherently flawed) polls!
> All he (or anyone) can do is interpret or analyse poll results, and then surface their findings in a way a larger audience can understand.
He (or anybody) can make adjustments to the data. He was challenged to explain why his predictions were so different, but he wouldn't do it.
> 538 did that better than any other poll analyst
He made a binary prediction, and it was wrong. There's no such thing as "better" when you only have one outcome. Your prediction is either right or wrong. If by "better" you mean he was wrong but assigned a higher probability to a Trump victory, the best forecaster would have been someone that mechanically changed the probability of a Trump victory to slightly less than 50% no matter what the data said.