From my view, 538 is the one cherry-picking data. I posted this in Poligaf, but it's probably worthwhile to post here too.
In June, 538 predicted trump would never go anywhere because his approval ratings were negative within his own party. Then something odd happened and his approval rating rose drastically alongside is rise in the polls. But they don't mention this change at all despite it once being their primary reason for writing him off. Instead they
made another post mid July, just this time saying his approval was below average instead of saying it was negative. But Trump's improving favorability didn't stop either, and now that Trump has mostly caught up to the other republicans, they don't even mention it anymore, never once admiting they were wrong about focusing on his favorables and assuming those favorables can't change. If that's not cherry-picking data, I don't know what is.
538 is right that he doesn't have enough of a lead where you can call the election right now, like you can for Hillary, but their evidence for saying he has a 0% is completely without statistical fact. They assume that people will learn to dislike him as time goes on, but there isn't statistical evidence to make that case, so they only make the case that there is still a chance he can lose, and say therefore he will lose.