Deep Hayer

analysis / sports

College basketball — 2025-26 model postmortem

#sports #modeling #postmortem

What the model got right, what it got wrong, and what I am changing for next season.

The 2025-26 college basketball season ended in April. Here is what my model did and what I am rebuilding before next November.

Headline numbers

What the model got right

The early-season conference play period. Schools where the roster turnover is high (transfer portal heavy) were systematically mispriced for the first two weeks of conference play. The model caught most of that.

Late-game close-and-cover situations were also a small but real edge. The market underprices defensive teams that play slow in close games; the model picks them up.

What it got wrong

NCAA tournament. The model treats tournament games as the regular season with a different schedule. They are not. The variance is higher, the rest-and-travel patterns are different, and the public bias is enormous. I lost 80 percent of my YTD CLV in the first weekend.

Player-prop integration was also a mistake. I added it midseason. It cannibalized attention from the moneyline work without adding measurable EV. Cutting it.

What I am rebuilding for next year

The rebuild starts in July. I will post weekly updates here once the new model is running in shadow mode.