How the Bracket Model Works
For now the bracket model is a simple placeholder. It ranks every team on published ratings, ESPN's BPI until NET arrives in December and then an average of the committee's team-sheet metrics, and builds the field in that order with the committee's procedure. A better model comes later. My own hand-built bracket is separate from it, and the record below is mine. NCAA Basketball
How it works
For now the model is a deliberately simple placeholder built from the committee's own metrics. Each team's rating averages its ranks in two groups, the predictive metrics (NET, KenPom, BPI and T-Rank) and the resume metrics (SOR, KPI and WAB), and weights the two groups equally. Before NET exists in December, the only full-field rating is ESPN's BPI, so the model uses that alone and says so.
From those ratings it builds the field the committee's way. The best-rated team in each conference stands in for its automatic bid, the best remaining teams take the at-large bids, and the seed lines fill in order, with the weakest at-large teams and the weakest automatic bids sent to the Opening Round. It does not yet place regions or keep conference teams apart.
Scoring uses the Paymon Scoring System, which grades exact seed placement and gives partial credit for proximity. Picking the right 68 teams and seeding them badly is a worse bracket than the selection accuracy suggests, and this metric catches that.
Note that scores are not comparable across formats. The 2026 bracket was built for the 68-team field with a maximum of 408 points. The 76-team era raises that maximum to 456.
Read how the committee builds the field
The record
Run on 2026's final ratings, the placeholder scores 295 of 408. Those ratings already include the NCAA tournament, so it only counts as a look back. The charts below are my own 2026 bracket, built before selection.
See the 2026 bracket scored line by line · Sixteen years, graded by Bracket Matrix
What it cannot see
The committee is twelve people in a room, and they weigh things no published metric carries: injuries and who was available for which losses, the eye test, conference politics, and the bracketing principles that pull teams off their true seed line for geography and conference separation.
The model is a placeholder with no live season yet. Its 2026 look-back runs on final ratings that already include the tournament, which flatters it, and my own bracket has one scored season on this page. Neither number is a track record yet.
The forecast is a seed list. It says nothing about who wins games once the field is set.
Women's Bracketology
Women's Bracketology is live in Bracket Studio, built on the same committee-style resume data as the men's field. There's no scored record here yet, because the model hasn't forecast a women's season. Once one wraps, its score joins this page the same way 2026 did for the men's field.