People do this by doing things like testing that their features contain information in both the training and test set. Because they are not exposing the data di

People do this by doing things like testing that their features contain information in both the training and test set. Because they are not exposing the data directly to the classifier they think that they haven't compromised the test set - but what they have done is increased the chances of a chance correlation.