Request for Clarification and Enhancement of K-Anonymity Counter Granularity

[leobierent12]

Subject:

Request for Clarification and Enhancement of K-Anonymity Counter Granularity

Issue Description:

Context:

On the specific page outlining the key parameters of K anonymity (p, w, k), and differential privacy parameters, we notice that the URL counter, denoted as C(S,t-w,t), must meet the k threshold to make the URL displayable. We seek to gain a deeper understanding of this counter.

Issue:

Currently, there is a significant gap in documentation concerning how the K-anonymity counter, C(S,t-w,t), is computed. We are particularly interested in knowing if it's possible to transfer counts from one device type to another. This aspect is crucial due to our use of renderURL/componentURL, which remains consistent across various device types and browsers as we display the same creative/products.

For instance, consider a scenario where we create a campaign with only one URL and target all devices (both app and web). The impact of the counter's granularity becomes evident:

If the counter is tracked by device type, separate counters would exist for app and web. This means that we need to surpass the K-anonymity threshold for both app and web (e.g., 50 users reached for each device), resulting in a total of 100 users required to display the URL.

Conversely, if the counter is cross-device type, we would only need to reach 50 users in total, regardless of the device type. This would be beneficial for browsers with lower traffic.

On our end, if the counter varies by browser x environment, the impacts on the share of clicks at risk by counter scope would vary significantly.

If we have only one counter, the impact of K anonymity will be way lower without variability by Device/browser. So, globally, we avoid any unexpected variation of K anonymity impact on perf depending on device and reduce sightly the impact on perf.

Question:

We need clarification on whether the K-anonymity counter will be consistent for a given URL across various:

Platforms/Countries

Browsers (e.g., Chrome Mobile, Chrome, Chromium)

Environments (Web, App)

Request:

At a minimum, we kindly request comprehensive documentation regarding the level of granularity at which the K-anonymity counter will be computed. Additionally, if feasible, we would appreciate efforts to maximize granularity or implement a solution that allows for a single counter across different device types, browsers, and countries. This enhancement would significantly improve our campaign optimization and delivery.

Your attention to this matter is greatly appreciated.

Labels: K-Anonymity, Documentation, Enhancement, Counter Granularity

[dmdabbs]

Great topic to bring to the weekly WICG PAAPI call:

[thegreatfatzby]

Hey @leobierent12 still worth getting an official answer from the G-Team but it could be worth reading the more detailed spec section 6 here, which gives natural language versions of the algorithms to be run when computing and incrementing k. Based on the listed algorithms I believe k counting should be independent of most of the variables you listed.

I suppose this would be broken if Google is planning to implement non-coordinating k-servers for different things like region, browser type, platform, etc. I don't see details of that in the spec (the detailed or less detailed one), so I think the question for @michaelkleber and team would boil down to whether there will be "one" k-server-cluster across all prod environments, that will be eventually consistent across regions, etc.

[michaelkleber]

The k-anonymity counter granularity only takes into account the ad render URL and the Interest Group's JS bidding/reporting function URL.

[leobierent12]

Ok, understood thanks @michaelkleber
The K anonymity counter will be computed no matter the region, browser type, platform...
Thanks for the clarification.
We can consider this issue as closed.