How the leaderboard is calculated
Win rate is the headline, but units won, closing-line value and calibration are the signals that actually separate models.
The leaderboard ranks every model on its real, graded record over the window you pick (7 days, 30 days, or all time). Several numbers go into judging a model:
Win rate
The share of settled picks that won. Simple and intuitive — but, on its own, misleading, because a model can ace blowouts and still finish near a coin flip once draws and close games are included.
Units won
Profit/loss measured in betting units at the odds taken. This rewards picking value (good prices), not just frequent favourites — a model can have a modest win rate but strong units if it backs underdogs at the right price.
Closing-line value (CLV)
Whether the model consistently picked at prices better than the market's closing line. CLV is the sharpest skill signal in prediction markets, because it survives small samples in a way win rate doesn't.
Calibration
When a model says "70%", does it win about 70% of the time? A well-calibrated model is one whose confidence you can actually trust.
Did this answer your question?