Model Lab
Every EdgeIQ model version is stored and graded against completed games. Historical predictions are never overwritten and weak results are published alongside strong ones.
EdgeIQ Baseline
Gradient-style baseline blending team power ratings, recent form, home advantage, rest and pace, with logistic probability calibration.
Sample
1390
Accuracy
69%
Brier
0.256
Log loss
1.129
Mean abs. error
27.14
Model performance detail
Month-by-month accuracy trends and the full calibration breakdown for this model version, so you can see how closely stated confidence matched real outcomes.
Your plan: Free · Unlocks with Pro — $9.99/month
Metrics are statistical measures of past performance on a limited sample. They describe how the model has behaved historically and do not guarantee future results.
EdgeIQ Baseline
Gradient-style baseline blending team power ratings, recent form, home advantage, rest and pace, with logistic probability calibration.
Sample
334
Accuracy
74%
Brier
0.191
Log loss
0.568
Mean abs. error
15.04
Model performance detail
Month-by-month accuracy trends and the full calibration breakdown for this model version, so you can see how closely stated confidence matched real outcomes.
Your plan: Free · Unlocks with Pro — $9.99/month
Metrics are statistical measures of past performance on a limited sample. They describe how the model has behaved historically and do not guarantee future results.
EdgeIQ Baseline
Gradient-style baseline blending team power ratings, recent form, home advantage, rest and pace, with logistic probability calibration.
Sample
2934
Accuracy
58%
Brier
0.273
Log loss
0.793
Mean abs. error
5.09
Model performance detail
Month-by-month accuracy trends and the full calibration breakdown for this model version, so you can see how closely stated confidence matched real outcomes.
Your plan: Free · Unlocks with Pro — $9.99/month
Metrics are statistical measures of past performance on a limited sample. They describe how the model has behaved historically and do not guarantee future results.
EdgeIQ Baseline
Gradient-style baseline blending team power ratings, recent form, home advantage, rest and pace, with logistic probability calibration.
Sample
1502
Accuracy
61%
Brier
0.237
Log loss
0.674
Mean abs. error
2.82
Model performance detail
Month-by-month accuracy trends and the full calibration breakdown for this model version, so you can see how closely stated confidence matched real outcomes.
Your plan: Free · Unlocks with Pro — $9.99/month
Metrics are statistical measures of past performance on a limited sample. They describe how the model has behaved historically and do not guarantee future results.
EdgeIQ Baseline
Gradient-style baseline blending team power ratings, recent form, home advantage, rest and pace, with logistic probability calibration.
Sample
3751
Accuracy
74%
Brier
0.186
Log loss
0.558
Mean abs. error
23.52
Model performance detail
Month-by-month accuracy trends and the full calibration breakdown for this model version, so you can see how closely stated confidence matched real outcomes.
Your plan: Free · Unlocks with Pro — $9.99/month
Metrics are statistical measures of past performance on a limited sample. They describe how the model has behaved historically and do not guarantee future results.
EdgeIQ Baseline
Gradient-style baseline blending team power ratings, recent form, home advantage, rest and pace, with logistic probability calibration.
Sample
6237
Accuracy
75%
Brier
0.168
Log loss
0.506
Mean abs. error
19.35
Model performance detail
Month-by-month accuracy trends and the full calibration breakdown for this model version, so you can see how closely stated confidence matched real outcomes.
Your plan: Free · Unlocks with Pro — $9.99/month
Metrics are statistical measures of past performance on a limited sample. They describe how the model has behaved historically and do not guarantee future results.
Community models
Models built by EdgeIQ members and published after review. Each one is graded on the same held-out games as the EdgeIQ baseline.
No community models have been published yet. Build one in the workbench and request publication.