Model Residuals
Calibration gives you one bias number per market. Residuals (actual − predicted) tell you where the model is systematically wrong. Mean residual < 0 means the model is over-predicting; > 0 means under-predicting. The histogram shows how the misses are distributed; the worst-misses table shows which exact plays broke the model.
Refreshed every morning by python/residuals.py after outcomes settle. Trailing 30-day window. See also Self-Learning for the bias correction that actually fires on tomorrow's projections.
Loading residuals...
Loading player patterns...
Loading park residuals...
Where does the model break per stadium? Parks with strong residuals suggest the model's park_factor multiplier is mis-calibrated for that venue.
Loading handedness residuals...
LHB vs RHB residuals. Skew in one direction suggests the model isn't accounting for platoon splits sharply enough.