Randy Ramsdell wrote: > IIRC, when I sent messages through sa-learn on the old mail server as > spam, then checking with spamassassin debug, this would show a 3.5 > BAYES score. I will double check this, but I would hope to at least > add a positive score when training a spam message.
Training a single message should not cause a large change in the Bayes score. Normally, you would need to train multiple similar messages before you see a large change. Keep in mind that Bayes works with statistics about the messages that it has learned. The change to the database from learning a single message depends on the tokens in that message and what Bayes had learned about those tokens previously. -- Bowie