A rule-based model predicting the 2026 NCAA Men's Basketball Tournament using power ratings derived from win percentage, scoring margin, and seed position.
View Full Bracket ↓rating = (win_pct × 30) + (PPG − OPPG) × 1.5 + (17 − seed) ×
0.8
The strongest factor. A team's sustained winning across a full season is the most reliable predictor. Weighted at 30 so mid-majors with elite records (e.g., High Point 30-4) are properly valued as upset threats against power-conference teams.
Points scored minus points allowed, amplified by 1.5× to reward dominant teams. Captures both offensive firepower and defensive strength. Houston's elite +13.5 margin generates 20.25 pts — more than many teams' seed bonus.
Intentionally reduced weight. Seeds encode committee judgment (SOS, eye test) but over-relying on them produces a chalk bracket. At 0.8, a 1-seed gets 12.8 pts while a 12-seed gets 4.0 — enough to be a tiebreaker without overriding strong stats.
No special upset rules are needed — the formula does the work. By weighting win percentage at 30 and reducing seed bonus to 0.8, teams like 12-seed High Point (30-4, rating 46.1) naturally outscore 5-seed Wisconsin (24-10, rating 40.2) because High Point's .882 win% generates a +5.3 advantage in the win% component, overcoming Wisconsin's +5.6 seed edge. The same mechanism lets 11-seed VCU (27-7) upset 6-seed North Carolina (24-8).
Hover over any team for detailed stats
Win percentage, scoring margin, and seed position represent three fundamentally different information sources: results (did you win?), performance (by how much?), and expert opinion (how does the committee rate you?). Using all three creates a balanced model that doesn't over-rely on any single metric.
Win percentage is weighted heaviest (up to 30 pts) because sustained winning over a full season is the best predictor. Scoring margin is amplified (1.5×) to separate dominant teams from close-game survivors. Seed bonus is reduced (0.8) so the data drives the bracket, not the committee's seed line. This means a mid-major with a great record can compete — just like real March Madness.
A pure seed model would predict 1-seeds always winning. But March Madness is famous for upsets. By reducing seed weight to 0.8 and boosting win percentage to 30, our model naturally predicts that 12-seed High Point (30-4) beats 5-seed Wisconsin (24-10), and 11-seed VCU (27-7) beats 6-seed UNC (24-8) — both based purely on the numbers, no randomness needed.
This model doesn't account for: injuries, matchup-specific advantages (size vs speed), momentum/hot streaks, strength of schedule beyond seeds, or coaching quality. A more sophisticated model would add adjusted efficiency metrics (like KenPom) and recent game weighting.