But, on the other side, nothing prevent you from modifying the client to
use a different or several servers for classification.
From the beginning, we wanted to let clients having their own
classification. The "official" one is just there as a reference point.
Yes, i have been thinking along those lines and i do think it is the way to go. Maybe we should split the current server into a core one and a server storing meta information (for example the urls associated with a response)? I think it is weird to store url for responses but not for the question by the way :)