All great points. Fangraphs seems to do a good job of recognizing error bars in the posts, chats, and affiliated podcasts they produce, so it is simplistic to take that "projections" page as a "predictions" page. Additionally, I would rather have the FG model than Fivethirtyeight's, which has shown the Brewers' playoff odds DOUBLE over the last week and a half. As nice as that is to see, it feels like pretty serious recency bias to me. This conversation gets to the heart of something I think is really important about analytics and projections. IMO, they still do a much better job of describing a change than predicting future performance. They are getting better at the latter, but the "error bars" SoCal referred to, the confidence intervals, they still tend to be pretty wide. Identifying a 95 percent range is much easier than noting where something will fall within that range.