After results are in, we go back to our own polling and check our work — where our model was right, where it missed, and why.