Transparency
Model Accuracy
Every prediction is checked against the real result. We score each fight using only what the model knew beforehand — a fair, out-of-sample test across 6,393 fights.
66.1%
On confident picks
when the model commits
59.3%
Every fight
incl. true coin-flips
0.236
Brier score
0.25 = a coin-flip
Calibration — the real test
Raw accuracy is the wrong yardstick in a sport this chaotic — what matters is honesty: a trustworthy model wins about as often as it says it will. When we call a fighter a 70% favourite, they should win ~70% of the time. Ours line up almost exactly:
50–60%
said 55% · won 55%
n=3878
60–70%
said 64% · won 64%
n=2180
70–80%
said 72% · won 80%
n=335