Brier score + skill

A Brier score grades probability predictions, like NovaPredict's boom% and bust%, against what actually happened. The companion skill number shows whether those probabilities beat simple guessing. NovaPredict publishes both as part of its public accuracy record.

The idea is simple. If Nova says a player has a 30% boom chance, that is a claim, and claims can be scored. When the week ends, the player either boomed or he did not. The Brier score rewards probabilities that leaned toward what happened and penalizes probabilities that leaned away, with confident wrong calls punished hardest. Say 90% and be wrong, and it costs you far more than saying 60% and being wrong.

How to Read It

Lower Brier is better, and skill is the number that gives it meaning.

• Brier score: a running grade across all of Nova's boom and bust calls. One player, one week tells you nothing; the accumulating record tells you plenty.
• Skill: the comparison against guessing. A naive forecaster could just assign every player the base rate and be "not terrible." Positive skill means Nova's probabilities are doing better than that; zero or negative means they are not adding anything. Skill is the honest test.
• What it means for you: positive skill over a real sample is your license to actually use boom% and bust% in lineup decisions, rather than treating them as decoration.

NovaPredict's live Brier and skill record starts in Week 1 of the 2026 season, built from projections locked before kickoff.

What It Is Not

A Brier score is not an accuracy percentage, and one busted 20% call is not a miss. If Nova's 20% boom calls hit about one time in five, those probabilities are doing their job perfectly. The score exists to judge the pattern of probabilities over time, not any single outcome.

For the score that grades the full range of outcomes rather than yes/no events, see CRPS.

Follow Brier and skill as the season fills in at /accountability.