Model Performance · v3.0 Walk-Forward Verified Out-of-Sample
Walk-Forward Accuracy
76.2%
76.2% walk-forward accuracy across 3,148 out-of-sample fights over 7 annual windows
Brier Score
0.1888
lower = better
Trend
↗ +2.12pp
per year
Per-Window Accuracy 1% — 1%
71
2020
70
2021
77
2022
76
2023
78
2024
80
2025
84
2026
149
ML Features
6,871
Training Fights
15K
Sims Per Fight
7
Validation Windows
Each window trains on strictly prior data and tests on that year's fights — zero lookahead. View per-bucket calibration →
Live Prediction Tracking (118 scored, 14 pending)
132
Total Predictions
118
Scored
76
Correct
64.4%
Live Accuracy
Accuracy by Confidence Bucket
Bucket Count Correct Accuracy
65%+ 31 21 68%
57-64% 15 9 60%
50-56% 79 50 63%
Recent Scored Predictions
Fighter A Fighter B Pick Confidence Actual Result
Justin Gaethje Ilia Topuria Topuria 76% Gaethje
Justin Gaethje Ilia Topuria Topuria 76% Gaethje
Ciryl Gane Alex Pereira Pereira 51% Gane
Ciryl Gane Alex Pereira Pereira 51% Gane
Aiemann Zahabi Sean O'Malley O'Malley 51% O'Malley
Aiemann Zahabi Sean O'Malley O'Malley 51% O'Malley
Derrick Lewis Josh Hokit Hokit 77% Hokit
Derrick Lewis Josh Hokit Hokit 77% Hokit
Michael Chandler Mauricio Ruffy Ruffy 76% Ruffy
Michael Chandler Mauricio Ruffy Ruffy 76% Ruffy
Kyle Daukaus Bo Nickal Nickal 54% Nickal
Kyle Daukaus Bo Nickal Nickal 54% Nickal
Steve Garcia Diego Lopes Garcia 54% Lopes
Steve Garcia Diego Lopes Garcia 54% Lopes
Belal Muhammad Gabriel Bonfim Muhammad 54% Bonfim
Model Edge vs Closing Line 47 predictions tracked
Model-quality metric. Edge = our model's confidence − closing implied probability (for our pick). Positive = we thought the pick was more likely than the market did at close. Different from CLV: this measures model disagreement with the market, not line movement.
-1.4%
Average Edge
17
+Edge picks
30
−Edge picks
36%
+Edge Rate
Edge by Confidence Zone
ZoneCount Avg Edge+Edge Rate
STRONG 8 -4.1% 38%
LEAN 11 +0.1% 36%
FLIP 28 -1.1% 36%
📚 View History by Event CLV Ledger Public Scoreboard Miss Archive 📐 Calibration Metrics