I have been using SB since 2203/4. The program is pretty simple in concept and use, but does not give much variety. You classify mail into two types, spam and non-spam, preferably over a longish period, before installing SB. Then you 'train' the program by feeding it the two types - it uses some AI rules to create some rules to use itself, and you provide two folders for SB to use - I created SpamForSure and SpamMaybe. You set up the parameters for these folders on a scale of 0 to 100. Mine are set so that a spam score of >20 will put an email in Maybe and >90 into ForSure. Nowadays, with automatic retraining, no 'good' messages get to either box. The auto retraining is implemented by hitting a Spam button which whips an offending email away and makes an update to the rules. As far as I can tell, this was probably an academic project for an AI student, and the result is very good, but it goes nowhere. There have been no updates forever and the observations on how it could be improved have been ignored. A pity. But it doesnt change the fact that the program works. Whether it is better than other filters around, I have no idea, that needs an unbiassed comparitive review.