In my business, I need to ensure zero false positives for spam, so I use magnets - heavily. It works well for me because my e-mail domain is mine and I give different e-mail addresses to different senders. It seems that most of the time spam occurs when someone sells my address or gets hacked. I mean how hard is it for a major software vendor to keep their customer database secure? Perhaps they don't care, because they'll never get caught, but when I get spam at <BigSoft?>info@<mydomain>.com, I pretty much know who leaked it.

Using this scheme, I can easily kill the address. If I still like the company, I edit my profile with them to give them a fresh address. I have an email client filter rule that immediately deletes all messages that have been "magnetted" with the spam-address magnet. For all others, I let the Bayes classifier do its thing and it does it well, even though it never sees the spam-address messages. I know the POPFile team discourages magnets, but that rubric is not necessarily for those who filter spam based on the "To:" header.

[Since my e-mail client does not show the "To:" header in its list of incoming mail (why should it, since it expects the reader to know their own e-mail address), I have hacked the Bayes.PM file to modify the Subject line when it suspects spam and adds the "To:" header to the Subject line so I can see what e-mail address the spam is coming in on.]

I guess I could hack bayes.pm to classify everything and THEN apply the magnet, but the classifier works well enough (scary well) with just the legitimate mail, plus the spam that has not been "magnetted" away, plus spam on the addresses I do not want to abandon.

However, the one thing I would really like to change in the magnet logic (either have someone else do it because I make a convincing case, or have me learn a little more about how the code is organized - Perl is not my first language - and submit that for approval) so that magnets can be exact matches or substrings, instead of just substring matching. For example, for the past 8 years or so, I have been getting spam at on@<mydomain>.com and I would like to kill it with a magnet, but since I have given out evolution@<mydomain>.com and creation@<mydomain>.com as my e-mail address and want to continue to use those, I cannot use a magnet to kill on@<mydomain>.com without killing all other mail that has a "To:" header that has on@<mydomain>.com as a substring.