NBAScopeStatistics • Predictions • AnalysisLIVE 7

How accurate is our NBA model?

We publish every miss. Predictions are frozen before tip-off, never edited, and scored after the final buzzer.

Live record — frozen predictions

27games scored
74.1%correct picks
0.173Brier score (0 = perfect, 0.25 = coin flip)

Frozen predictions started on November 12, 2026, so this sample is still small. The backtest below shows the long-run picture.

Backtest, 2016-17 to 2025-26

Each past game is predicted only with the ratings available before it — no hindsight. Over 12,852 games the model picks the winner 64.9% of the time with a Brier score of 0.218. The baseline “always pick the home team” is shown for comparison.

SeasonGamesCorrect picksBrier scoreHome team won
2025-26132268.4%0.20855.5%
2024-25132165.7%0.21454.5%
2023-24131965.3%0.21454.7%
2022-23132063.3%0.22958.1%
2021-22132364.8%0.22354.8%
2020-21117161.6%0.22954.7%
2019-20114364.1%0.22154.7%
2018-19131264.6%0.21459.0%
2017-18131265.6%0.21458.7%
2016-17130964.9%0.21858.2%

Calibration — when we say 70%, does it happen 70% of the time?

All 13,049 backtest games grouped by the favorite's probability. A well-calibrated model has the two columns close together.

Favorite's probabilityGamesPredictedFavorite actually wonGap
50–60%4,33055.0%54.4%-0.7 pts
60–70%3,97864.8%62.9%-1.9 pts
70–80%2,87574.6%71.6%-3.1 pts
80–90%1,65484.2%82.6%-1.6 pts
90–100%21291.9%88.7%-3.3 pts

Reading: in every band favorites won less often than predicted, by 0.7 to 3.3 points — the model is somewhat overconfident.

Biggest misses since November 12, 2026

Model details in the methodology; current team ratings on the Elo ratings page. Probabilities are estimates, not betting advice.

Cite this page

You are welcome to quote these figures. Please credit NBAScope with a link to this page.

NBAScope, “How accurate is our NBA prediction model?”, https://nbascope.com/nba/predictions/accuracy.