Over on SFF.net in one of the tech help groups they are discussing the sff.net spam filters. The discussion reminded me all over again why I’m glad I stopped using any of their filters and chose to do it all myself, via SpamAssassin, Razor, Popfile, CRM114, et al. I love SFF.net, and they are great guys who run a great system. However, their spam stance is unacceptable to me. I would rather delete 1000 spams than miss a mission critical e-mail. When I hear some of the choices they make, it makes me shudder.
Because I use it for training Bayesian filter, I have a corpus of all my spam since last November, as well as snapshots of non-spam taken regularly over that time (but not comprehensive.) They mentioned in this discussion they block all mail from Topica. I checked my spam corpus and in over 6000 spams I had 22 from topica. Meanwhile, in my corpus of 2200 righteous e-mail that I have, there are 46 from Topica. For me, a Topica e-mail is at least 6 times more likely to be legitimate. This is but one of many heuristics they use that just are too nonspecific for my taste. True, spam is a huge problem for them, but their particular solution is in my view worse than the problem. That’s why I’ve been handling it on my own and been much happier for it.