bloom filters

[patmmccann]

would it make sense to return a bloom filter instead of cohort id's? cohorts could then be tested against the bloom filter to see if the user were in the particular cohort with some chance of false positive (no chance of a false negative). Also the publisher might be blind to the advertiser cohort of interest and could just pass along the bloom filter.

[brodrigu]

This would provide additional privacy benefits such as not needing to share browsing history in order to place a user in a given cohort.

However, to truly realize those benefits we would have to remove the ability for the cohort provider to analyze traffic patterns to generate the model / bloom filter. This may make it a less effective feature for advertising, but this might ultimately be a necessary trade-off.