Scores, schedules and lines from ESPN; past seasons from sportsdataverse. Not affiliated with any team, league or bookmaker.
Scorecard
How the model would have done on past seasons, predicting every game the morning it was played using only earlier games, and how it is doing this season.
Updated Thu 5 Nov, 4:27 PM ET
Men's
Picks right, past seasons73.0%70,802 games, 2013-14 to 2025-26
Log loss0.523Lower is better; 0.693 is a coin flip
Average margin error8.8 pointsPredicted against final margin
This season85.3%128 of 150 picks, 2 of 3 logged before tip-off
Picks right by season
Every game held out
Are the chances honest?
Home win chance against how often the home team won
Against the betting market
2,983 games from 2024-25 and 2025-26 with a closing line
Picks right
Log loss
Margin error
Total error
This model
72.8%
0.519
9.1
13.2
Closing line
73.8%
0.512
9.0
12.8
An even mix of the two
—
0.513
—
—
The closing line is the sharper forecast: it picks 1.0 points more games right and its log loss is 0.008 lower. Mixing the model into the line does not improve it, so the line already contains what the model knows.
By part of the season
Past seasons
Games
Picks right
Log loss
November
13,398
77.2%
0.460
December
13,126
76.6%
0.469
January to March
44,195
70.6%
0.559
NCAA tournament
801
71.0%
0.541
November is easiest: many games pit strong teams against weak ones. Conference play and the tournament match teams of similar strength.
Season by season
Each season scored with settings fitted on the others
Season
Games
Picks right
Log loss
Brier score
Margin error
Total points error
2025-26
5,752
72.0%
0.521
0.177
9.1
13.8
2024-25
5,764
72.9%
0.519
0.176
8.9
13.0
2023-24
5,726
72.3%
0.538
0.182
9.0
13.2
2022-23
5,721
71.8%
0.540
0.182
8.9
13.3
2021-22
5,503
72.5%
0.525
0.177
8.8
13.5
2020-21
3,868
71.1%
0.543
0.186
9.3
13.4
2019-20
5,328
72.8%
0.527
0.177
8.9
13.4
2018-19
5,603
73.6%
0.520
0.175
8.9
13.6
2017-18
5,540
74.2%
0.516
0.173
8.8
13.7
2016-17
5,521
73.2%
0.519
0.175
8.7
13.6
2015-16
5,469
73.7%
0.513
0.173
8.6
13.9
2014-15
5,499
73.6%
0.517
0.174
8.4
13.4
2013-14
5,508
74.3%
0.511
0.171
8.3
13.7
Women's
Picks right, past seasons76.9%41,259 games, 2018-19 to 2025-26
Log loss0.466Lower is better; 0.693 is a coin flip
Average margin error9.6 pointsPredicted against final margin
This season82.6%119 of 144 picks, 5 of 6 logged before tip-off
Picks right by season
Every game held out
Are the chances honest?
Home win chance against how often the home team won
Against the betting market
916 games from 2024-25 and 2025-26 with a closing line
Picks right
Log loss
Margin error
Total error
This model
82.2%
0.372
9.8
13.4
Closing line
82.6%
0.369
9.4
12.6
An even mix of the two
—
0.365
—
—
The closing line is the sharper forecast: it picks 0.4 points more games right and its log loss is 0.003 lower. An even mix of the two (0.365) beats the line alone, so the model knows something the line misses; on 916 games that is suggestive, not proof.
By part of the season
Past seasons
Games
Picks right
Log loss
November
7,880
79.5%
0.419
December
7,205
79.3%
0.419
January to March
26,131
75.4%
0.493
NCAA tournament
461
80.3%
0.403
November is easiest: many games pit strong teams against weak ones. Conference play and the tournament match teams of similar strength.
Season by season
Each season scored with settings fitted on the others