Model calibration audit

For every bucket of predicted probability, what did the model actually hit? A well-calibrated model has actual win rate ≈ predicted prob. A negative gap means the model is overconfident in that bucket (dangerous); positive gap means it's underpredicting (safe). Rows turn red when |gap| > 5pp AND N ≥ 10.

Calibration cohort by sport (click to filter):

All sports MLB N=2,748 NCAAF N=1,323 TENNIS_WTA N=746 TENNIS_ATP N=652 MMA_MIXED_MARTIAL_ARTS N=518 NHL N=451 LALIGA N=274 AMERICANFOOTBALL_NFL N=271 SERIEA N=226 EPL N=224 LIGUE1 N=195 BUNDESLIGA N=179 UCL N=158 NFL_PRESEASON N=145 CRICKET_IPL N=27 NBA N=9

Combined "all sports" is rarely meaningful — sports differ in market efficiency, signal availability, and base rates. Use the chips to drill into a single sport. N<50 (red) means the calibration is brittle; N≥200 (green) is trustworthy.

Filter: window=90d · sport=tennis_atp

Overall: N = 652 · mean predicted 64.8% · actual win rate 66.3% · gap +1.4pp · Brier 0.216 · log-loss 0.622

Calibration by predicted-probability bucket.
Predicted-prob bucket	N	Mean predicted	Actual win rate	Gap (actual − predicted)	Brier
50-55%	94	51.9%	58.5%	+6.7pp	0.246
55-60%	202	57.8%	60.4%	+2.6pp	0.240
60-65%	82	61.8%	62.2%	+0.4pp	0.234
65-70%	83	67.8%	68.7%	+0.9pp	0.212
70-75%	51	71.4%	60.8%	-10.6pp	0.248
75-80%	83	77.5%	80.7%	+3.2pp	0.158
80-90%	37	84.1%	86.5%	+2.4pp	0.115
90%+	20	92.1%	85.0%	-7.1pp	0.133

JSON feed · Model CLV · Methodology

Calibration is computed from the prediction_history table. Bucket edges are 5pp wide (50-55%, 55-60%, etc.) with widened tails (<50%, 80-90%, 90%+) where data is sparser. Negative gap is the dangerous direction — actual win rate below stated confidence.