📊 Model Calibration Dashboard - 2025
How well do our predicted probabilities match actual outcomes?
Calibration by Probability Bin
Do our 60% calls hit ~60% of the time?
No calibration data available. Run the backfill command to generate synthetic props.
Understanding Calibration
- Calibration Gap < 3%: Excellent calibration
- Brier Score: Measures probability accuracy (0 = perfect, 1 = worst). Lower is better.
- Synthetic Props: Free calibration testing using season averages as lines
- Real Props: Tracked against actual bookmaker odds (costs API credits)
- Sample Size: Need 20+ props per bin for reliable calibration