Is your feature request related to a problem? Please describe.
In segmentation (personalization) use cases, we need to report on how each defined audience performs relative to the others. Currently, we do not have any way to track this data. This used to be tracked under the audience checkpoint, but got removed for privacy concerns.

Describe the solution you'd like
We want to re-introduce the checkpoint, and we want to be able to track this in a privacy compliant way, as discussed over slack.
Differential audiences is what we settled upon where we will randomly select whether we report the "true" audience, or pick a random one from the list. And in the reporting, we'll compute the variance from a uniform distribution to estimate the audience distribution.