I use rspamd for spamfiltering. But I’m not happy with it’s work.
I’ve never had any false-positives, and rarely false negatives, but since rspamd both happen very often. (and false positives are the worst, since my (family) users don’t check their spam-folder too often.)
ofcourse, I try to teach rspamd by moving mail to and from spam-folder.
In the docs I read:
The Bayesian tests **are not active until it has received enough information. This includes a minimum of 200 spams AND 200 hams (false positives).
The rspamd webinterface shows me a counter of learned mails, but not specified to spams and hams. Is there a way to check if I already meet the 200+200 requirement?
And will the learning make a big difference? Or is it just a very minor adjustment?
The common feedback is the opposite: rspamd works quite well.
Anyway, bayesian filters will adapt to your emails, it may prove really helpful.
You could use rspamc stat to see how many spam and ham has been learned (see the last two lines).
You could also look at wrongly classified emails headers to see if some rules fire frequently.
Finally, not the easier tool, rspamd_stats </var/log/maillog could help to understand what’s happening.