running model performance statistics disaggregated by race — is an essential step before deploying any predictive model in political targeting contexts. Bias discovered before deployment is correctable; bias discovered after is a harm already done.