Hey all, I have a mail server set up using spamassassin for a company which has several mail forms. When spam started getting sent through the forms, and several form emails were (correctly) marked spam, those receiving the form mails were suddenly concerned that they might eventually lose a good email from the form. In the spirit of "we'd rather receive 500 spam mails than lose one sales lead", the VP suggested that we simply add a rule where we add -50 to the spam score if its subject contains the subject written by the form. (That way all future replies to the form mail are exempt from ever being counted as spam, too.) Obviously, this solves any potential form mail loss.
But what does it do to the Bayesian auto-learn? Suddenly we could have form spam coming through with ridiculously negative scores. I have autolearn set to learn messages with scores of -5 as ham. Obviously if I read all of the form mail, I could manually relearn any form spam as spam. Or if I could actually convince the sales people to put it in a special mailbox....but I doubt our salespeople would remember. Is there a better way of doing this? Renee -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean.
