What a football model has to prove

I love sports. The football-analysis project gives that interest another outlet: looking closely at the difference between a convincing story and a pattern that holds up.

The project has explored a backtest on the 2024 NFL season. The questions matter as much as the results: what information was available before kickoff? What would a simple baseline have predicted? Does a promising result hold up across more games?

I'm treating this as a learning project, not a prediction service. A good explanation of why a model was wrong can be more useful than a confident explanation after a win.