It appears that many (or most) people use SA with Bayes, although it looks like a few are not. We are currently opting not to use Bayes.
For those who are not using Bayes: 1. Why not? 2. What are you doing above and beyond default SA config? We are: 1. Adding most all the custom rulesets: backhair, tripwire, antidrug, plus some others seen here. 2. Tweaking the scores for the default SA rules (I know about the "genetic algorithm, but some make no sense at all - rules that score 4.3 that hit twice as much ham as spam and others that score .1 that hit almost only spam.) 3. Adding a few custom rules. 4. Parsing mail log to MYSQL database to analyze SA rules and scores for SPAM and ham for future rule adjusting. 5. Threshold = 10.0 The biggest problem we've had for FPs has been being too aggressive on the blocklist scores (SORBS, NJABL, DYNABLOCK). We've gotten so many FP's with those rules that we've actually added some meta rules to lower the score for email that hit those rules in combination. We just started SA sitewide - about 350 users - and we are getting very few FP's. 1 a day or less being reported to us. We are getting false negatives reported - probably about 10 a day. Anybody doing something similar? Al
