This by design, and is appropriate.   

User complaints are unreliable, so user feedback affects only that one user.   Abstractions of that feedback are passed back to Spamfoo, and used to modify future revisions of their model.  They have to separate noise from truth as they make those decisions.

To get a domain-level or system-level filtering system, you need to invoke Spmfoo with a user id that represents the domain admin or system admin.  This requires a small amount of custom coding.  Then you need an admin who is qualified and willing to do the work of making those decisions.   He will need a good message review tool that helps him find the needles in the  haystack of all received messages.

I have assembled a workflow to do this.   I run Spamfoo only on my incoming gateway, and only on the most uncertain messages.  The results are reviewed carefully and then feedback is used to maintain a system-level personalization.  Honestly, it is a little ugly, but it is working.

A general outline of my approach is here:

Send me a private message if you want to move forward.