Hey, thanks for a great post! Regarding the user learning/novelty effects, do you suggest any methods to account for them (other than running a long experiment)?