On 08/31/2012 04:56 PM, Timothy Murphy wrote: > It appears to be running OK, but only catches about 20% of my spam. If you have old rules and as far as spam evolves, that situation is "fair" > I wonder if others have experienced this decline? No. > I suspect that sa-learn is not performing properly. > (I run sa-learn nightly on my spam, > which is saved in ~/Maildir/.Spam/cur/ , > through the default spamassassin installation.) - how old is the Spam to feed sa-learn with? - what weight do you give to bayesean detection? > Eg I receive more or less identical spam many times > without it being caught. > > Also, about 35% of the spam that gets through > is in foreign languates - Russian, Chinese, Arabic and Turkish - Those are local langages for some... > One difficultly I have is that there seem many places > where one can make changes in spamassassin settings. > Am I right in thinking that all such changes can be made > in .spamassassin/user_prefs ? Seems OK > What should I put in this file to change the score for eg > 3.5 BAYES_99 BODY: Bayes spam probability is 99 to 100% Advices: - SA can be asked to report the scores in one header: make it reports and analyze the scoring - SA has a mailingg list you could also trigger in order to help -- RMA.