I may be missing something, but if you train a model to pass a specific test… isn’t it obvious that it would do better on that test?

I thought we called models with test data in their training set “poisoned”