The quality demands of machine learning are steep, and bad data can rear its ugly head twice both in the historical data used to train the predictive model and in the new data used by that model ...