my concern about this is if a user saw two variations, how would you know which variation causes the change in the metric?

practically, you can add any attribute as a an 'identifier' and then you can add that as the hashAttribute ("Assign value based on attribute") value

when setting up the feature flag experiment