Prediction model leaderboard

Every trained model, ranked by how far it beats the benchmark its own market has to clear. Accuracy across different markets is not comparable, because a line that falls one way most of the time hands a model a high score for learning very little. The gap to the benchmark is what ranks here.

3 of 4 beat their baseline by more than a point3 on a provisional sample (under 1,000 matches)

Sort byEdge over baseline Accuracy Log loss Sample size

SportAll football basketball tennis

Models ranked by edge over their market baseline
#	Model	Edge	Accuracy	Baseline	Log loss	Sample	Since last train
1	NBA winner basketball · Moneyline · Always pick the home side	+9.02	66.14%	57.12%	0.6154	7,012	▲ 1.25
2	NBA total 225.5provisional basketball · Totals · Always pick the majority class	+5.20	60.34%	55.14%	0.6629	827	▲ 0.53
3	NBA total 235.5provisional basketball · Totals · Always pick the majority class	+1.08

Edge is accuracy minus the baseline, in percentage points, measured on the held-out test period each model was evaluated against, never on data it trained on. A model with no recorded baseline shows its accuracy but no edge, because there is nothing to say how much of that accuracy is skill.