Predictions

When we say 60%, teams win about 60% of the time. Scored in the open — every pick locked in before the match, then scored against the result.

Calibration check · dot size = sample size
0.2233 calibration score
63.8% winner called right

Lower is better — 0.250 is a coin flip. Every match scored was one the model never trained on.

coin flip 0.250 always favourite 0.238 ratings-only 0.231 Calibrated 0.2233
4,069 matches scored · all unseen in training

Upcoming

locked in before each match

Loading the upcoming slate…

Past / Results

scored against the result

Every past pick is shown — wins and losses. What we're judged on is calibration, not raw accuracy: a 56% call that loses isn't a broken model — it's a 56% call.

Loading results…