2026-09-17

How the model works, in one page

Every team gets a single rating: how many points better than an average Division I team it is on a neutral floor.

Solving every game at once

A game's margin is modelled as the home team's rating, minus the away team's, plus home-court advantage. With about 6,000 regular-season games and 360 teams, that's a big system of equations. We solve it all at once (a ridge regression, which keeps teams with few games from getting extreme ratings). The result is automatically adjusted for schedule: beating a top-20 team by 5 moves your rating more than beating a bottom-20 team by 20.

From ratings to odds

A rating gap becomes a win probability through a curve fitted on the last stretch of the regular season. A 10-point gap is roughly an 85/15 game.

What we tried and dropped

We tested adding KenPom, Barttorvik, EvanMiya, betting lines, seed history, injuries, coaching, player-level models and even a transformer on every possession since 2021. Over many seasons none of them reliably beat the plain rating on tournament games; the noise in one 63-game tournament is simply bigger than their effect. So the site runs on our own numbers only.

How good is it?

Level with the best public systems: ahead of all three over the last three tournaments, behind over the longer record, and never by enough to call it a real difference. The full scorecard is on the Simulations page.