In article <[EMAIL PROTECTED]>, [EMAIL PROTECTED] said... > > Hello, > The documentation for 2.60-rc3 claims the bayes auto learn cutoff values > are -2 & 15. I think they have been changed to 0.1 and 12. Is that > correct?
There is certainly something odd happening here, I've been seeing a fair few spam mails that score sometimes as many as 50 points but have autolearn=no in the X-Spam-Status header.
Since I use the autolearn=spam tag as a way to bin spam that I don't need to feed through sa-learn, I'm puzzled.
Anyone know why this happens. FWIW I sometimes see two nearly identical spams (where maybe they were addressed to different users) that score the same however one is autolearned and the other isn't.
Something odd here. Using 2.60-rc3 BTW on RH 7.3
Doesn't sound odd to me. If they're nearly identical are you sure that they don't have the same message id ?
Also, if you get a lot of mail comming in at once then autolearning has to lock the bayes database and the timeout on the locks is quite short (deliberately) so lock contention will cause some messages not to be learnt.
There are quite a number of reason for autolearning not to learn something, its not just a matter of if its greater than 15 learn it....
Regards, Simon
------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
