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=2,486 NCAAF N=1,487 TENNIS_WTA N=734 TENNIS_ATP N=659 NHL N=591 MMA_MIXED_MARTIAL_ARTS N=526 AMERICANFOOTBALL_NFL N=313 LALIGA N=302 SERIEA N=276 EPL N=253 LIGUE1 N=222 BUNDESLIGA N=206 NBA N=166 UCL N=158 NFL_PRESEASON N=145 CRICKET_IPL N=27

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=nhl

Overall: N = 591 · mean predicted 49.9% · actual win rate 49.6% · gap -0.3pp · Brier 0.235 · log-loss 0.663

Calibration by predicted-probability bucket.
Predicted-prob bucket	N	Mean predicted	Actual win rate	Gap (actual − predicted)	Brier
<50%	252	40.4%	37.7%	-2.7pp	0.232
50-55%	176	52.4%	54.5%	+2.1pp	0.248
55-60%	75	57.4%	50.7%	-6.7pp	0.254
60-65%	44	62.4%	75.0%	+12.6pp	0.202
65-70%	32	66.7%	62.5%	-4.2pp	0.237
70-75%	9	72.1%	100.0%	+27.9pp	0.078
75-80%	3	76.1%	66.7%	-9.4pp	0.230

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.