I gave each one a full month across two cities, which made the differences much easier to see. Consistently, how much the interface pushes a subscription drives the outcome more than luck does. It held steady reliably after I controlled for the obvious variables.