All Sports Performance

AD TEST MODE

PERFORMANCE HEADER

Mock Placement 728x90

College Basketball (M)

basketball • Performance Deep Dive

Trained Never Predicted Never

All-Time Average (28 folds)

Flat ROI

+100.0%

Hit Rate

63.6%

+11.2pp vs b/e

Recent Trend (last 3 folds · Feb 12–Mar 26, 2026 ) ↑ Improving

Flat ROI

+112.9%

+12.9pp vs avg

Hit Rate

61.8%

Out-of-sample Accuracy

68.2%

Avg across 28 folds

AUC-ROC

0.542

0.5 = coin flip · 1.0 = perfect

Log Loss

0.6088

Lower is better

Brier Score

0.2105

Probability calibration

28 validation folds

Feb 1, 2024 – Mar 26, 2026

Final Holdout Test

Held back from all training and fold evaluation — the cleanest out-of-sample signal.

No holdout data available. Re-run prediction-system run --sport mens-college-basketball to generate holdout metrics.

Fold-by-Fold Metrics

Supporter+ required

Supporter access required

Unlock fold-level charts to identify training stability.

Unlock Supporter

Betting Simulation

Supporter+

Recommendation ROI

�

Recommendation Win Rate

�

Universe ROI

�

Universe Win Rate

�

Supporter access required

Unlock Supporter

Model: LightGBMModel

Calibration: platt

All-Time Average (28 folds)

Flat ROI

+19.8%

Hit Rate

61.7%

+9.3pp vs b/e

Recent Trend (last 2 folds · Feb 12–Mar 26, 2026 ) ↓ Declining

Flat ROI

+14.5%

-5.3pp vs avg

Hit Rate

58.8%

Out-of-sample Accuracy

55.9%

Avg across 28 folds

AUC-ROC

0.517

0.5 = coin flip · 1.0 = perfect

Log Loss

0.6771

Lower is better

Brier Score

0.2422

Probability calibration

28 validation folds

Feb 1, 2024 – Mar 26, 2026

Final Holdout Test

Held back from all training and fold evaluation — the cleanest out-of-sample signal.

No holdout data available. Re-run prediction-system run --sport mens-college-basketball to generate holdout metrics.

Fold-by-Fold Metrics

Supporter+ required

Supporter access required

Unlock fold-level charts to identify training stability.

Unlock Supporter

Betting Simulation

Supporter+

Recommendation ROI

�

Recommendation Win Rate

�

Universe ROI

�

Universe Win Rate

�

Supporter access required

Unlock Supporter

Model: LightGBMModel

Calibration: platt

All-Time Average (28 folds)

Flat ROI

+12.5%

Hit Rate

58.5%

+6.1pp vs b/e

Recent Trend (last 2 folds · Feb 12–Mar 26, 2026 ) ↑ Improving

Flat ROI

+15.4%

+2.9pp vs avg

Hit Rate

60.3%

Out-of-sample Accuracy

55.7%

Avg across 28 folds

AUC-ROC

0.516

0.5 = coin flip · 1.0 = perfect

Log Loss

0.6751

Lower is better

Brier Score

0.2414

Probability calibration

28 validation folds

Feb 1, 2024 – Mar 26, 2026

Final Holdout Test

Held back from all training and fold evaluation — the cleanest out-of-sample signal.

No holdout data available. Re-run prediction-system run --sport mens-college-basketball to generate holdout metrics.

Fold-by-Fold Metrics

Supporter+ required

Supporter access required

Unlock fold-level charts to identify training stability.

Unlock Supporter

Betting Simulation

Supporter+

Recommendation ROI

�

Recommendation Win Rate

�

Universe ROI

�

Universe Win Rate

�

Supporter access required

Unlock Supporter

Model: LightGBMModel

Calibration: platt