On Mon, Oct 05, 2009 at 10:45:40AM +0100, Ned Slider wrote: > Mike Cardwell wrote: >> >> I use SpamHaus from SpamAssassin rather than directly from my MTA >> mainly because I don't want that mail to avoid the bayes auto-learning. >> If I ever find the service running out of cpu cycles, I might consider >> putting the SpamHaus checks before SpamAssassin, but until then I may >> as well use the resources I have available. >> > > You don't need to have everything bypass smtp level checks to do that - > a couple good spam trap addresses that are whitelisted should be good > enough to feed Bayes, but yes, it's good to have a full spectrum of spam > for Bayes to feed on.
It's really hard to make assumptions about Bayes. One _could_ assume that most directly blockable zen/botnet/etc stuff has very little resemblance on the spam that bypasses all that (freemail, 419 etc). YMMV, but my own stats would support that. Bayes works so well here that I fail to see why I'd need to learn all that botnet stuff.
