With almost all races called, we have a LONG review up this AM on how the 538 forecast did last month. The short answer is: really good! "Called" 95-97% of races right. Almost exactly predicted House seat count and popular vote. Pretty good on Senate/Gov. 53eig.ht/2SwymKR
But we also have a number of self-critiques/areas-to-improve. A fun one—if you care about calibration—is that *not enough underdogs* won. e.g. Candidates in "likely" races were supposed to win 86% of the time but actually won 94%. This may or may not (see the story) be a problem.
Also have some thoughts about election night, where our House forecast shifted toward the GOP quite a bit at one point in the evening before snapping back to Democrats. There are things for us to look at here, but I also think election night models are REALLY intrinsically hard. 
But overall this a technical story because we think the PHILOSOPHICAL debates on election models—and to a lesser extent, "data journalism" overall—are dumb. This stuff has been around for a long time now. It has a good track record. It deserves a role in coverage. It works.
