🏆 Bracket Simulator

Interactive 2026 NCAA Tournament bracket based on our ML model predictions. Download the prediction files to submit to the competition.

How to Submit

The full prediction files (every possible matchup) are what you submit to the competition — not just the bracket view below.

1
📥
Download Files
Download both MNCAATourneyPredictions.csv and WNCAATourneyPredictions.csv
2
✅
Verify Format
M file = 72,010 rows, W file = 71,253 rows. Columns: WTeamID, LTeamID
3
📧
Submit to Captain
Email/share both files with your team captain before March 17 at Noon EDT
4
🔄
Progressive Updates
For progressive brackets, new predictions can be generated after each round using real results
📥 Men's Predictions (MNCAATourneyPredictions.csv) 📥 Women's Predictions (WNCAATourneyPredictions.csv)
File format: Each row = one predicted matchup outcome. WTeamID is the predicted winner, LTeamID is the predicted loser. All C(N, 2) pairs are listed — not just the actual bracket matchups.
📥 Download Predictions
—
Total Matchups Predicted
68
Tournament Teams
196
Max Possible Score
6
Rounds
How to read this bracket: Teams are shown with their seed number (lower seed = stronger team). The green highlighted team in each matchup is the model's predicted winner. Confidence % shows the model's estimated win probability.

Note: This visualization uses seed-based simulated matchups. The full prediction CSV files contain all 72,010+ real men's predictions and 71,253+ women's predictions based on actual team statistics.

Bracket Types

📋 Regular Bracket

All predictions are locked in before the tournament starts based on seedings and season statistics. Winners of later rounds are predicted based on who we expect to win earlier rounds.

Risk: upsets in early rounds cascade through the bracket, potentially costing many points in later rounds.

⚡ Progressive Bracket

After each round, actual winners are plugged into the model to generate new predictions for the next round. This allows the model to adapt to Cinderella stories and upsets.

Advantage: errors don't cascade — a Round 1 upset won't destroy all your later predictions.

Historical Backtest

Simulated bracket scores on past tournament seasons. Models were trained only on prior seasons' data — no data leakage.

SeasonCorrect GamesTotal GamesAccuracyBracket ScoreScore / Max (196)
Loading...
These scores are from synthetic data. With real Kaggle NCAA data, expect bracket scores of 80–130+ out of 196 (top Kaggle competitors reach 140+).