This bit is crucial for running a good experiment: > Good experiments use tight exposure groups Not all your users will have the conditions necessary for the

This bit is crucial for running a good experiment: > Good experiments use tight exposure groups Not all your users will have the conditions necessary for the experimental treatment. You should only look at changes in behavior where the treatment condition was true. Otherwise your experimental effect is diluted. But your comparison between experiment and control groups must be neutral on the condition, that is, users in the control group would have seen the treatment if they were in the experiment group (counterfactual).