Re: [docs-and-reports] Private measurement of single events (#41)

Thanks for filing this issue Charlie, and I'm glad to see it's going to be a discussion topic at our next meeting.

I guess the big question is: **Do we believe a "Private Measurement API" must necessarily include aggregation?**
- PCM does _not_ necessarily include aggregation, but the total number of individuals for whom you can query event level data is just 256 per website, which makes it economically unattractive to try to use the API for this purpose.

I would *really* love to hear what Mozilla, Brave, and Webkit think about this topic. CC: @martinthomson, @ShivanKaul, @johnwilander

If we think the answer is "Yes, a private measurement API _must_ necessarily include aggregation", we should discuss what mechanisms we could use to try to enforce that.
1. There is the PCM approach, where you limit the total number of breakdowns, such that there is an upper bound on how many single events can be measured.
2. You can enforce a minimum threshold of at least K inputs having a given breakdown key. If an adversary can generate K-1 fake events, this doesn't make it _technically impossible_ to measure single events, but it makes it cost K-times as much to do so.
3. You can enforce a minimum threshold of at least K events which _contribute to the output_ for a given breakdown key. The same logic applies about adversaries who can generate fake events, with the added concern that this will make reporting effectively useless for small advertisers, who often have small budgets with only a handful of total attributed conversions.
4. We can just make sure that we chose epsilon such that the signal-to-noise ratio if terrible you _attempt_ to measure a single event, effectively making it unattractive to even attempt to do so.

That's all the ideas that immediately come to mind, but I might be missing something. Happy to hear other ideas from other people!

-- 
GitHub Notification of comment by benjaminsavage
Please view or discuss this issue at https://github.com/patcg/docs-and-reports/issues/41#issuecomment-1517069569 using your GitHub account


-- 
Sent via github-notify-ml as configured in https://github.com/w3c/github-notify-ml-config

Received on Thursday, 20 April 2023 23:52:14 UTC