Does anyone know the experiment targeting rules for user attributes that change value? e.g. • If a user doesn't initially match the target attributes - the first time they would otherwise be in the experiment - can or should they be included in the experiment if they match later on? • Once a user has matched attributes and been included, do they continue to see experiment variants if the targeting conditions subsequently fail? I can see in general it's a lot of complexity having changing user attributes although in some cases it could be quite useful

Experiment assignment is evaluated in realtime, so that if targeting attributes change, then a user could be added/dropped from an experiment variation. Sticky bucketing offers an alternate pattern that can allow a user to persist in an experiment assignment.