On Fri, 16 May 2014 16:20:21 -0400 Bowie Bailey <bowie_bai...@buc.com> wrote:
> Keep in mind that BAYES_50 and BAYES_60 still contribute positive > scores by default. Though it is technically a neutral result, it > still adds a point or two to the score. > Rather than messing with Bayes, I would focus on the spams you are > seeing and try to find a common thread that you can use to make a > custom rule or two to catch them. If they all have similar garbage > appended to them, there are probably other similarities you could > find. I have already made many such custom rules. As I wrote, that's mostly what I was working on this week :-( For instance, I noticed many of them (but not all) put my address in the Message-ID. Some (but not all) use broken HTML template kits that leave nice fingerprint marks in the body. And so on. But usually only 1 of them fires, at most - that is a 1.0 score, BAYES_50 is also around 1.0 I think, and that's about it - no RBL hits, no Razor or Pyzor hits. And to add insult to injury they almost always hit RP_MATCHES_RCVD, for a (locally modified) -0.15 boost. So, these rules are helping, but not enough. I am still getting about 1 unkilled spam an hour, which is too much for me. Today I have enabled full auto-learning (prior to this, I had bayes_auto_learn_on_error = 1). Hopefully that will give Bayes much more learning material. -- Please *no* private copies of mailing list or newsgroup messages.