From an earlier thread, it seems that a handful of us are having trouble with bayes working against us to negatively (as in negative score) tag obvious spam. For me, it's to the point that more than half of my spam gets through because almost everything gets BAYES_00=-2.5. If there was a magic bullet for figuring this out, I'm sure someone would have posted it by now, but is there at least process we can use to try to figure out what is mis-training the bayes filter. I didn't have this problem on a standalone SA/Amavis install, so I'm assuming something is significantly different about the process Zimbra is using..