@evan Have they considered doing it in smaller stages with subsets of users, for example only users in a certain language or geography? That would also allow to pay closer attention to the most affected federated instances, instead of having to observe effects scattered everywhere all at once.