WICG / turtledove

TURTLEDOVE
https://wicg.github.io/turtledove/
Other
537 stars 237 forks source link

Reporting on User signals and trusted server signals #213

Open BasileLeparmentier opened 3 years ago

BasileLeparmentier commented 3 years ago

Dear Michael,

As already discussed in meetings, and pushed by Jonasz in this pull request https://github.com/WICG/turtledove/pull/194 I would like to reiterate the need to have reporting on both the trusted server signals and user bidding signals.

This reporting may be differential private, but it is extremely important to have reporting on signals that are used in the bidding if we want to be able to optimize our campaigns, leading to higher ROAS for marketer and spend on publishers.

Without this, we won't be able to use fledge as intended during the origin trials and the testing steps, meaning that we will have misleading results to share, as either we will use models not trained on fledge (not available once 3rd party cookies are removed) or used a degraded version of fledge without the signals that can be used to have better performance.

Could you please therefore confirm that you intend to incorporate reporting on all signals in fledge, by a secure reporting mechanism, as soon as possible in the process?

Many thanks!

michaelkleber commented 3 years ago

Are you asking about the ability to do aggregate reporting from inside the Fledge worklets, i.e. tying these signals to auction outcomes? Or is this about a use of conversion measurement, tying these signals to later user behavior on a different site?

BasileLeparmentier commented 3 years ago

I mean both. It is important for both use cases, we need positive and negatives, to have reporting on the cost of the auction for instance, and then conversion reporting in order to be able to optimize ML models

vincent-grosbois commented 3 years ago

There is also another upside if reporting on trusted signal is allowed at event level : this would allow us to more easily sign reporting events. For instance: