The tracking poll is garbage but ignoring that. Do you actually think a one point movement in two consecutive polls from a pollster constitutes a trendzline?
The Upshot hasn't shifted from the 84 it declined to after a speight of tighter but still Clinton lead polls. It didn't dip below 84 afaik. So I don't know where you're getting improving chances from.
What the 538 national polling tracker shows is that Trump tightened the race by about 4 whole points from late october into the beginning of November. since then, the race has been pretty flat, and a poll drop today showed maybe just over a half a point gain in Clinton's favor from the day before. Now you can't really call this solid trend yet, since really we're looking at a reversal of movement within a span of basically a single day, but we already saw that the Trump rise trend that caused 538 to dip into the 60% range for Clinton did stop.
My point here is that there is some subtle movement in the polls that would cause a more volatile model like Silver's to show a slight upswing in Clinton's direction. RCP's polling average graph is showing a much more dramatic reversal today. I want to try and dispel the notion that Silver is necessarily manually poking his model back to Clinton as a response to criticism rather than his model just being his model.
I have no idea if Silver's model is perfect or not. Honestly all of the models are making unscientific assumptions to some extent. And even when models are wrong they can be wrong in very subtle ways. And we have no possible way to check for the correctness of polling and prediction models within the middle of a race because the entire population votes only once.
More than anything, you might see me "defend" 538 here more often than other posters is because 1) I don't think Nate being a bad pundit really has any bearing on whether or not his prediction model is good or not and 2) some people in PoliGAF seems very eager to dismiss polls or methodologies that do not favor Clinton. If anybody thinks they're immune from confirmation bias, they are dead wrong.