Every Division I game, every day, rated before tip-off and published next to the sportsbook line, so you can see where the model disagrees with the market, and how often it has been right to.
No 2025–26 game had been predicted live yet when this was built, so this is a replay: the current model, run on each game using only data available before tip-off. The live record replaces it once this season's games are graded.
Most public predictions are a number with no history attached. This site is the opposite: every prediction is stored before tip-off, locked once the game starts, and graded after it.
The goal isn't to beat the bookmakers on every game — nothing does. It's to find the games where the model and the market genuinely disagree, and to be honest about the rest.
Every game on the predictions page is one row of the scoreboard. This is a real graded replay — Oklahoma vs Colorado — with each column numbered.
99
Oklahoma
vs
99
Colorado
Before the game: the tip-off time in Eastern time, or TBD. LIVE blinks while it's being played, and FINAL shows once it's over.
Away team on top, home team below with an @, or vs at a neutral site. Before the game the side the model picks carries an amber bar and a PICK tag; once it's played, the bar moves to whoever actually won and the score windows light up.
The last sportsbook spread before tip-off, favourite first: −6.5 means that team is favoured by 6.5. Before the game it shows the Edge, the gap between the model's number and the market's; afterwards ✓ ATS, ✕ ATS or = Push.
The biggest move at any one book since it opened. Green means the market drifted toward the side the model backs; red means it moved against it. The column is left off days when no line has moved.
Written in the same notation as the Vegas line, so the two compare directly:
Oklahoma −6.5 means the model has Oklahoma winning by 6.5.
Next to it, that side's calibrated win probability. replay marks a prediction rebuilt after the season rather than published live.
✓ when the model picked the winner, ✕ when it didn't, and how many points it missed the final margin by.
The › at the end of each row opens the matchup page: both teams' recent results and every sportsbook's line. This row is only an illustration.
Box-score efficiency for every Division I team: points per possession on both ends, shooting, turnovers, rebounding and free throws, rolled over the season so far, plus an Elo rating that carries form from game to game.
A gradient-boosted model predicts the final margin from the two teams' profiles and where the game is played. A win probability turns that margin into odds, and it is calibrated: on last season's games, when it said 70%, that side won about 70% of the time.
In November a team has played a handful of games, so early predictions lean on a preseason prior built from last season and the preseason polls. That prior fades as real games accumulate. Treat the opening weeks with some skepticism: the model's error is largest there.
Each prediction is shown beside the sportsbook spread, with how far the line has moved since it opened. Finished games are graded against the line they closed at, so the record against the spread is a number rather than a claim.