I am performing manual review of items sent to the Training folder.    My customized external sender warning lets the user know that it is avaiable:

I do not have the option of automatically feeding a spam filter's Bayesian model, but I don't think I would dare.   Manual review of the feedback is helping me block illegitimate and unwanted mail, but it also includes stuff that should be handled differently.  This includes:

SmarterMail has an integral :"Unsubcribe" button which is displayed for messages that it can parse correctly.    I have not decided how to help users notice that it is there and safe to use.   That capability is related to the largeer question of helping users know whether an unsubscribe link is safe or not.   So for the moment, I am doing a lot of unsubscribe that users could do for themselves.

When I do block messages, I have to dig deep to determine which identifier should be blocked.  The attacker's identity could be hidden by impersonation or hidden in the Reply-To field, among other possibilities.    In some cases, the attacker's identity is fully obfuscated and the only available response is an Abuse report to the attacker's hosting service (if that service is legitimate.)

True there are also the no-reply@yourdomain.com sent from googlecontentusers. These tend to get through also. We wouldn't want user inedvertenly blocking their own domain.