Hi ! I'm using SpamAssassin 2.60 with autolearning. Today I received one of the usual penis pill advertisments, which triggered: | X-Spam-Flag: YES | X-Spam-Checker-Version: SpamAssassin 2.60 (1.212-2003-09-23-exp) on | cslmail.nrw.net | X-Spam-Report: | * 5.4 BAYES_99 BODY: Bayesian spam probability is 99 to 100% | * [score: 0.9989] | * 0.1 HTML_MESSAGE BODY: HTML included in message | X-Spam-Status: Yes, hits=5.5 required=5.0 tests=BAYES_99,HTML_MESSAGE | autolearn=ham version=2.60 So it was marked as Spam due to the Bayesian filter. But as autolearning happens before the Bayes score is added, and scoreset 1 is used (which gives HTML_MESSAGE only 0.001), the score is below the default bayes_auto_learn_threshold_nonspam (0.1), and the message is autolearned as Ham ! I've checked the same message after this autolearn: now the Bayes score dropped to 0.56, so it wasn't marked as Spam anymore...
So this isn't really a bug (SA works exactly as documented), but I wonder if this is the intended behaviour. IMHO a message marked as Spam should never be autolearned as Ham. Thomas -- Thomas Mechtersheimer [EMAIL PROTECTED] ------------------------------------------------------- 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
