Overfitting means a model has learned the noise in its training data, not the real signal. It performs beautifully on old matches and poorly on new ones.

It happens when a model is too complex for the amount of data, or when a builder keeps tweaking it until a test set looks good.

Warning signs include a huge gap between training and test performance, rules that only make sense for one team or season, and endless feature additions.

The defenses are simple: prefer simpler models, test on later data, limit tuning, and demand consistent performance across seasons. Set limits and never bet what you cannot afford to lose.

See the final section of the full article for additional practical guidance on this topic.