When I was a student, I used to spend weeks tweaking hyperparameters to improve my models for Kaggle competitions by 1 or 2%. Not once did I look at the data to see what my model was doing wrong.

Share your shame.