one thing i haven't been able to understand?: if each cohort group is so small (relatively i think in the thousands) combine with a UA should be 100% unique?
even if cohort is in the millions a UA+ip or geo should be enough to ID, or even add a couple more bits of window.property entropy enough to stay under the 'budget' limit
even if cohort is in the millions a UA+ip or geo should be enough to ID, or even add a couple more bits of window.property entropy enough to stay under the 'budget' limit