Total Games
4,813
Unique games backtested
Best Accuracy
65.6%
formula 40 + elo 60
Correct Picks
3,157
of 4,813 games
H2H Boost
+0.8%
head-to-head blending
Rolling Accuracy — Best Model NBA
Best model: Formula 40% + Elo 60% (65.6% accuracy). Head-to-head blending adds ~+0.8% on top. Formula weights: win %, home/away splits, last-N form, rest days, injuries.
Winner Accuracy — Model Comparison NBA
ModelAccuracyCorrectGamesConfidence
formula 40 + elo 60 ★65.6%31574813
formula 60 + elo 4065.6%31554813
formula 50 + elo 5065.4%31504813
elo only65.4%31464813
Total Games
1,085
4 seasons (2022–2025)
Regular Season
1,032
Weeks 1–18
Playoff Games
53
Wild card → Super Bowl
Best Accuracy
65.4%
rec30 + adv30 + elo40
Rolling Accuracy — Best Model NFL
Regular season best model: 30% Record + 30% Advanced Stats + 40% Elo (65.4%). Advanced stats = yards/play, turnover margin, 3rd-down %. Home field: +2.5 pts / 48 Elo pts.
Playoff finding: Pure Elo (75%) beats every blend — regular-season advanced stats don't translate to playoffs. Model auto-switches to Elo-only in playoff mode.
Regular Season — Winner Accuracy NFL
ModelAccuracyCorrectGamesConfidence
rec30+adv30+elo40 ★65.4%6681021
rec33+adv33+elo3365.1%6651021
rec40+adv20+elo4065.1%6651021
rec25+adv25+elo5064.9%6631021
elo only63.4%6471021
record only63.0%6431021
Total Games
2,740
Unique games backtested
Best Accuracy
71.0%
rec50 + elo50 N=6 K=32
Correct Picks
1,945
of 2,740 games
Home Field
+4.0
pts / 55 Elo advantage
Rolling Accuracy — Best Model CFB
Best model: 50% Record + 50% Elo, N=6 recent games, K=32 Elo factor, home field +4.0 pts / 55 Elo pts (71.0% accuracy across 2,740 games).
Winner Accuracy — Model Comparison CFB
ModelAccuracyCorrectGamesConfidence
rec50+elo50 N=6 K=32 ★71.0%19452740
rec50+elo50 N=6 K=2071.0%19442740
rec50+elo50 N=6 K=2870.9%19422740
rec50+elo50 N=6 K=2570.8%19392740
Total Games
7,402
3 seasons (2023–2025)
Best Accuracy
56.7%
rec30 + elo70 · N=40 K=4
Correct Picks
4,196
of 7,402 games
Home Field
+25
Elo pts advantage
Best model: 30% Record / Run-Differential + 70% Elo, N=40 recent games, K=4 Elo factor (56.7% across 7,402 games). Baseball is the hardest of the four sports to predict — outcomes are high-variance, so even sharp models live in the high-50s.
Winner Accuracy — Model Comparison MLB
ModelAccuracyCorrectGamesConfidence
rec30+elo70 N=40 K=4 ★56.7%41967402
rec30+elo70 N=40 K=656.7%41957402
rec40+elo60 N=40 K=456.6%41937402
home team only (naive baseline)~54%—7402
Total Bouts
1,710
3 seasons (2023–2025)
Best Accuracy
55.1%
fighter Elo · all bouts
Decided Bouts
1,681
used in backtest
Fighters
1,054
tracked by Elo
v1 baseline: per-fighter Elo (career win-rate blend didn't improve it). The honest signal is ~55–57% on the full set — high-accuracy rows with very few bouts are small-sample noise, not skill. Vegas favorites win ~57–65%, so the gap is what fighter stats (reach, finish rate, takedown defense, layoff) would close — a planned upgrade.
Winner Accuracy — Model Comparison UFC
Fighter Elo — all bouts ★55.1%9261681
Fighter Elo — ≥1 prior fight56.6%5741015
Fighter Elo — ≥5 prior fights (small sample)69.4%2536
Model Learning History
Loading…
Create a Matchup
Run the model on a live game, an upcoming game, or any two teams you pick — even a matchup that isn't scheduled. Powered by the same engine behind our accuracy stats (NFL & College Football).
VS