SpamAssassin uses the same scoring system for Bayesian filtering as for other rules. Rather than using a multiplier, where each percentage point adds a fixed amount to the score, the probabilities are banded, and the bands assigned individual scores. This allows for greater flexibility in scoring.
In the default SpamAssassin configuration, a number of levels or bands are identified, and each is assigned a different score. The default score values are relatively conservative—from 0% to 60% the scores are negligible and beyond that the bands are assigned an increasingly high score. This approach helps to avoid false-positives. There is no direct relationship between the Bayesian probability of an email being spam and the value that is added to SpamAssassin's score for the email as a result of that probability.
The scores for Bayesian rules can be edited as for other rules. Bayesian rules have names such as BAYES_00
and BAYES_10
. The exact rules for a particular version can be retrieved...