[tbrownaw]

What were the error rates for the various groups with the old process? Was the new process that included the model actually worse for any group, or was it just uneven in how much better it was?