> thinking about some data > come up with some crazy edge cases where our data would be incorrect > spend entire day fixing it to handle these properly > finally run the code > data accuracy improves by 0.03% why do i do this to myself