In a contest like this, I would expect (and I could be wrong) that there is a final validation set not accessible until "game day". Thus, overfitting to a train

In a contest like this, I would expect (and I could be wrong) that there is a final validation set not accessible until "game day". Thus, overfitting to a training data set of any kind would be detrimental and degrade model performance.