Some time this week, we will be making changes to our SpamAssassin configs to use a site-wide Bayesian learner instead of per-user. Given that the vast majority of our spam is directed towards our mailing lists rather than individual club members, I think this makes the most sense for us. We will also be increasing the scores for the Bayes tests to keep spam out of our inboxes more aggressively.
Unfortunately, this means that there will likely be some false positives. I will try to adjust the Bayes test scores to keep this rate low. Alternatively, you can set your own custom scores in ~/.spamassassin/user_prefs.
This means that all of the bayes* files in everyone's ~/.spamassassin directories are now useless. If these files are taking up a lot of space, it would probably be a good idea to delete them.
As a final reminder, SpamAssassin only adds headers to emails which it flags as spam - it does not actually delete or move them. If you wish to invoke some custom action on these flagged messages, you can use a custom procmailrc file, as outlined in our wiki: https://wiki.csclub.uwaterloo.ca/Mail#Spamfiltering
If you have any concerns or suggestions, please send a message to email@example.com.
-- Max Erenberg Systems Administrator | Computer Science Club University of Waterloo