Performance

How the models have performed

Real outcomes, not selected highlights. Live pregame predictions and historical chronological tests are separate records; every rate below has a sample size.

Live predictions graded after games

EdgeIQ saves projections before a game and grades them against the final result. Live grading is different from a historical test: it reflects predictions made in operation rather than unseen games held out during model development. Rates need at least 30 completed predictions; a dash means that threshold has not been met.

Predictions graded

175

Overall win accuracy

66% across 175 graded predictions

By league

Live pregame winner projections only
LeagueGradedWin accuracy
NFL29—
NBA0—
MLB5—
NHL0—
NCAA Football14166%
NCAA Basketball0—

Historical chronological testing

For each league, the current engine is evaluated on later games it did not see during training. These numbers are backtests, not live prediction results. A test accuracy rate cannot be combined with a live accuracy rate.

NFL · version v3.2.0

62% winner accuracy over 780 held-out games

Average margin error: 10.7 points over 780 tests

Average total-score error: 10.5 points over 780 tests

NBA · version v3.2.0

67% winner accuracy over 524 held-out games

Average margin error: 11.6 points over 524 tests

Average total-score error: 15.2 points over 524 tests

MLB · version v3.2.0

55% winner accuracy over 1162 held-out games

Average margin error: 3.5 points over 1162 tests

Average total-score error: 3.4 points over 1162 tests

NHL · version v3.2.0

60% winner accuracy over 568 held-out games

Average margin error: 2.2 points over 568 tests

Average total-score error: 1.9 points over 568 tests

NCAA Football · version v3.2.0

68% winner accuracy over 1028 held-out games

Average margin error: 16.3 points over 1028 tests

Average total-score error: 13.0 points over 1028 tests

NCAA Basketball · version v3.2.0

70% winner accuracy over 1948 held-out games

Average margin error: 8.8 points over 1948 tests

Average total-score error: 13.3 points over 1948 tests

Probability quality and score error

The archived model record below is a separate dataset from the current engine tests above. Brier score compares each predicted probability with the actual binary result; lower is better. Average score error is shown only where enough predictions have been graded. Model versions may overlap, so these rows must not be added together.

EdgeIQ Baseline · vv1.0.0 · NBA

1390 graded · Brier: 0.256 · Average score error: 27.14

EdgeIQ Baseline · vv1.0.0 · MLB

2981 graded · Brier: 0.272 · Average score error: 5.08

EdgeIQ Baseline · vv1.0.0 · NHL

1502 graded · Brier: 0.237 · Average score error: 2.82

EdgeIQ Baseline · vv1.0.0 · NCAA Football

4406 graded · Brier: 0.176 · Average score error: 23.55

EdgeIQ Baseline · vv1.0.0 · NCAA Basketball

6237 graded · Brier: 0.168 · Average score error: 19.35

EdgeIQ NFL Performance · vv2.0.0 · NFL

1966 graded · Brier: 0.234 · Average score error: 15.55

By confidence range

Confidence is a model label, not a guarantee. Accuracy for small groups is withheld to avoid making a handful of games look conclusive.

LeagueConfidenceGradedWin accuracy
MLBhigh183861%
MLBlow54750%
MLBmedium59656%
NBAhigh120872%
NBAlow9653%
NBAmedium8653%
NCAA Basketballhigh439682%
NCAA Basketballlow73349%
NCAA Basketballmedium110861%
NCAA Footballhigh137394%
NCAA Footballlow153955%
NCAA Footballmedium149480%
NFLhigh28770%
NFLlow91855%
NFLmedium76165%
NHLhigh76367%
NHLlow35447%
NHLmedium38560%

Recent performance

A recent-window rate is not displayed until it can be calculated from enough timestamped, graded pregame predictions for that window. Not enough graded predictions yet. Historical version-level trends are available in the Model Lab where entitled.

Past performance does not guarantee future predictive accuracy. Learn how these measures are produced in our methodology, or read the analytics disclaimer.