Over/Under bets showed strong performance (3-0) - The model's total predictions were well-calibrated, particularly the WAS@BOS Under 232.5 which hit by 21.5 points
Injury impact calculations were directionally correct - Heavily injured teams (SAC, WAS) did struggle offensively as predicted, leading to successful Under bets
WHAT MISSED
Spread predictions completely failed (0-3) - All three spread picks lost, with the "HIGH confidence" BOS -20.5 missing by 9.5 points despite a 23.5 fair value calculation
Overestimated injury impact on spreads - The model's injury adjustments (+3.5 for WAS, -10 cap for SAC) weren't nearly enough to account for actual performance gaps
Home court advantage assumptions broken - LAC lost straight up at home as 13.5-point favorites to an injury-decimated SAC team
MODEL CALIBRATION NOTES
Confidence ratings were poorly calibrated - The "HIGH confidence" BOS pick was the worst performer, suggesting overconfidence in BPI-based calculations
Clear systematic bias toward favorites - All three spread losses were on the favorite side, indicating the model undervalues underdog variance and motivation factors
METHODOLOGY ADJUSTMENTS TO CONSIDER
Reduce injury impact calculations - Current injury adjustments appear 2-3x too large; consider capping total team injury impact at 5-6 points rather than 10+
Add variance buffers for large spreads - Spreads above 10 points show higher miss rates; require larger edges (4+ points) before recommending rather than current 3-point threshold