On Tue, March 31, 2009 18:35, Karsten Bräckelmann wrote: > Uhm, Benny, you do realize that there appears to be no manual ham > training according to the OP, do you?
only trusted senders learn bayes ham here, but all can learn spam, end result is that it maintains good here the op might have the problem in not let sa autolearn good enoug on score based learning, but on the manual learning some mails as spam that is not really spam ? also the otherway around is bad learning msgs that gives sa scores of -100 in spamassassin -t msg, bayes will not hold it forever so whitelist_spf and whitelist_dkim just not on domain only senders whitelist_spf postmas...@example.com def_whitelist_spf *...@example.com adj scores on them def_ should not give much, just enough to not make msgs quarantined, and thats it -- http://localhost/ 100% uptime and 100% mirrored :)
