How to read these numbers

The card says Seattle 62%. That is not a promise Seattle wins. It says that in the 100 weeks that look like this one, Seattle wins about 62 of them and loses about 38. If you only look at our 62% calls, roughly four in ten should go the other way. When that stops being true, the number is wrong, and the track record shows it.

A model that says 55% for every game, home team every time, gets more than half its calls right and tells you nothing. A model that says 90% when it means 90% and 52% when it means 52% is useful even when its call rate looks similar. So we score how close the stated chance is to the observed rate, not how many calls landed.

Take the stated chance, subtract what happened (1 for a win, 0 for a loss), square it, and average over every call. Zero is perfect. A coin flip scores 0.25. Anything near 0.20 is elite. Lower is better, and the difference between 0.22 and 0.21 is a season of work.

The closing line is the best public forecast that exists; it has every model and every bettor priced into it. No published model beats it over a season, and we do not expect to. Our target is to sit close behind it, with game ranges that contain the truth about nine times in ten, and a player's floor to ceiling about six in ten. A final game's page shows the closing line beside our call.