Model Calibration
Edge vs Actual Outcome
Measures whether the model's edge predictions match reality. If model predicts +20% edge on a pick, that pick should win 20% more often than its Kalshi price suggests. Drift from diagonal = miscalibration.
Total Settled
3
3W - 0L
Actual Win Rate
100.0%
3 of 3
Avg Predicted Edge
+16.0%
model average
Calibration Gap
138.1%
underconfident
Prediction Accuracy
Each point = one pick. X = model's predicted probability. Y = actual outcome (win/loss).
ROI by Edge Bucket
Actual ROI per edge range. Higher edge should yield higher ROI.
How to read this
✓ Well-calibrated: actual win rate per bucket matches predicted probability. Scatter points land on the diagonal.
⚠ Overconfident: model predicts 80% but wins 60%. Reduce stake sizing or raise edge threshold.
○ Underconfident: model predicts 60% but wins 75%. Can increase sizing — there's more edge than model claims.