The new Bayesian filtering seems to work well but there is still the nagging bit of how to get those few pieces of spam into the corpus so they can be scanned for in the future.

It has been suggested to offer tha ability to hold all mail for an admin to sift through.  To me this would be a ridiculous chore to have to go through and the holdup on getting urgent email would anger a lot of users.

The problem seems to be how to keep the email format from modification by the end users email application.  My idea is to have a seperate database to store a COPY of all passed email.  This can be set with an aging period much like the regular quarantine to keep it from getting too large.  When spam passes through undetected the admin can go to the gui and find the copy and select to add it to the corpus as spam.  Since this copy is still in the correct format it would not cause any problems with SF.   In the process this would also remove the incorrect "good" counts (that were tallied for the original email) from the corpus. 

What do you think?

>>My idea is to have a seperate database to
>>store a COPY of all passed email. 

That was certainly what I was thinking as well.

I doubt that anyone wants the personally vet all the good mail on an individual basis...