Status: Draft.

Objective

Predict the win probability for every possible NCAA tournament matchup.

Approach

The gap between two teams' Massey rankings already says most of what I need to know. So the core model is small:

1. A logistic map from Massey rank difference to win probability.
2. Massey features across ranking systems.
3. Bookmaker-odds normalisation: remove the bookmaker's margin from the odds so they can be compared with my probabilities directly.

The XGBoost and Elo notebooks are kept as references only. I'm not using them in the submission.