How well does this actually work?
Every number on this page is computed from stored predictions, not typed in.
Three kinds of evidence, ordered by how hard each is to fake. The first is what every draft tool publishes. The second is what almost none do. The third is predictions made before the game was played, scored against what happened -- and it is the only one that cannot be produced by choosing a favourable split.
Loading...
AUC is the probability the model ranks a random winning draft above a random losing one; 0.5 is a coin flip. Pre-game prediction has a low ceiling by nature -- ten players and thirty minutes of gameplay happen after the draft ends -- so the number to compare across tools is not how high it is, but whether it was measured on games the model had never seen. ECE is average calibration error: how far the predicted probability sits from the observed win rate, which is what decides whether "56%" means anything at all.