GORD is our ranking of how strong each team's case for the NCAA tournament is, and it picks the 76 teams on the bracketology page.
How It Works
Machine-learning models learned from past seasons which teams make the tournament, using stats like adjusted offense and defense and strength of schedule.
Men: two models, one on Bart Torvik's T-Rank stats and one on KenPom's, make up 80% of the score; the T-Rank, KenPom, NET and ESPN BPI rankings add 5% each. Women: 50% model, 30% T-Rank, 20% NET. Through Dec 1 only the rankings count; the models grow to these weights by Feb 1.
The field: conference champions (each conference's top team until its tournament is played), then the best of the rest, to 76. Each group's 12 lowest seeds play the Opening Round.
The seeds follow the same order: GORD #1 is the top 1 seed, four places to a line.
On the scoreboard: GORD #N is a team's place in that order; NET is the NCAA's own ranking; WAB is wins beyond what a bubble team would have against the same schedule. Seed is the AP ranking until the bracket is out, then the tournament seed.
Keep in Mind
It ranks a rรฉsumรฉ, not who would win on a neutral floor, and the committee's bracketing rules aren't modeled.
When It Updates
Once a day in season, Nov 1 to Apr 10.
For the Curious
Each model is a stacked ensemble: several different models whose answers another model combines. They learned on full-season records, so in November they read a 1-0 start as a perfect season: hence the wait. NET and BPI join once they're released for the season; until then the other parts share their weight. The Opening Round (12 games; the First Four until 2027) pairs champions with champions and the rest with each other, neighbors in GORD order. The separate Power Rankings page is Torvik's T-Rank, averaged with BPI once it's current.