On Wed, 05 Aug 2009 10:15:00 +0200 a...@exys.org wrote:
> 2 to 5 is the sweetspot. That message in question actually proved it > is working, since the URIBL hits came later. Then it scores >10 so > it gets rejected. I noticed earlier that you were greylisting for only 60s; that seems like a fairly short delay to affect listing. > I think that setup is fairly smart, I don't run my own mta, but I do something analogous, in that I do an initial test with Bogofilter and use the result to delay spam up to 24 hours before it's processed with SA. I think if I were doing greylisting I might use Bogofilter's ham result to bypass it, and the unsure/spam results to set short or long delay. > excluding the problem that i train SA with wrong information. I think if Bayes is being mistrained, you have the autotrain thresholds wrong. And in your situation, it's not going to be possible to reverse it properly since the signature will change with the received headers.
