13 keys really strikes me as overfitting past data and expecting it to hold when all you did was fit noise (with malleable subjective keys, too). Doesn't help that number of keys doesn't seem to be correlated at all to margin of victory.
But his track record is exemplary so I don't think we can write it off.
He was wrong in 2000. And I don't want to hear any "popular vote" stuff because if his keys predict that, then it was wrong in 2016.
He has "predicted" 10 elections with one failure. 90% is cool and all, but it's not like these are coin flips.