BetPick AI
Model Transparency

How our model actually performs

Predicted probability vs. actual hit rate across 646 graded picks and 11 markets. Updated daily. No cherry-picking, no removed dates.

Last refresh: 7/15/2026, 8:30:03 AM

Calibration Curve

A perfectly calibrated model sits on the diagonal — 60% picks win 60% of the time.

65-70%
51 picksActual: 60.8% · Δ-6.7pp
70-75%
198 picksActual: 61.1% · Δ-11.4pp
75-80%
112 picksActual: 65.2% · Δ-12.3pp
80-85%
102 picksActual: 64.7% · Δ-17.8pp
85-90%
125 picksActual: 72.0% · Δ-15.5pp
90-95%
58 picksActual: 94.8% · Δ+2.3pp

Brier Scores by Market

Lower is better. 0.25 = pure chance, <0.20 = strong edge. Weight = how much we trust the market line vs. our model.

SportMarketBrierMarket WeightSamples
Not enough graded samples yet.
Data source: pick_calibration and pick_market_weight — refreshed nightly from graded picks.