What goes in
- Team ratings. Each team has a rating: how many points better or worse than an average team it is. After every game a team's rating moves by a small share of how far the result beat or missed what we expected. A new season starts from part of last season's rating, since rosters change.
- Home court. Home teams win by a few points more than they otherwise would. Games on a neutral floor (the league's games abroad, the NBA Cup final) get no home edge.
- Tired legs. A team playing the second night of a back-to-back loses a little.
- Who is missing. Each regular player is worth some points, from his share of his team's minutes and how much better his box score is than a bench player's over the last 15 games. A player ruled out takes his full value off his team; a questionable player takes off his value times our chance he sits. A traded player counts only for his new team.
The predicted margin becomes a chance with a bell curve: a 7-point favorite wins about two games in three. During a game the same sum runs with the real score and the time left, so a lead late counts for more than the same lead early.
What it cannot see
- Lineup news posted after the league's last injury report.
- How two teams match up, beyond their ratings.
- Motivation: a team resting stars late in the season, or playing for a seed.
- During a game, who has the ball and how many fouls each team has.
How good it is
It was tuned on 2021-22 to 2024-25 and tested on 2025-26, which played no part in tuning it. On that test it gave the favorite 67% on average, and the favorite won 68% of 1,310 games. Every chance shown before tip-off and during games since November 20, 2026 is scored too. See the record.