Claude Frantz wrote: > Hello Folk ! > > Sometimes I get BAYES_99=3.5 as result of the test, in a message > coming to me. This message is only sent to me and its contents is not > SPAM. > > How can I find exactly the reason on this erroneous marking ? The general reason is the message closely matches your spam training.
To find the exact reasons (there will be many, bayes is token based), feed the message to spamassassin with bayes debuging enabled: spamassassin -D bayes <message.txt It should show you a list of every token matched and its bayes probability as a range from 0 to 1.0. It's normal for ham messages to match a few 0.99's, but they'll match many more 0.01's. In this case, it would appear to be matching many more 0.99's than 0.01's. Note: MAKE *ABSOLUTELY SURE* to run the above test as the correct user. ie: the user whose bayes db gets used when you scan mail. (This is very often NOT the To: user. You're looking for the user passed after the -u parameter, or the user that executes spamc, which might be a system account depending on how you've set things up)
