Anyway I’ll see how it goes. If it doesn’t seem effective enough then I can always add other things. Thanks for the suggestion, I’ll consider policyd-weight as a backup option if my current configuration seems inadequate.

You probably don’t want to accept spam from dodgy IP address X, because they happened to include some key phrase in the spam (say “Debian”), which your filters have previously regarded as “almost sacrosanct”.

Hence sometimes you can make a better decision using information that can’t be faked or is expensive to fake/forge, versus considering both that data, and data that is produced by the spammer.

Perhaps it is the varied interests of our clients, but I find the error prone nature of content filtering frustrating.

Worse yet a certain distro’s mailing list server is very fussy and seems to get annoyed when you reject too much of the spam it forwards. That led to trouble when I weakened, and allowed some content based filtering, and rejected emails that other people’s Spam Assassin instances had already tagged as spam (Incoming messages that claim to be spam often are!). I should have guessed those admins weren’t too bright, spending CPU resource to find it is spam and sending it anyway. So rejecting or delaying email from mostly good servers may cause other issues.

Our most spammed server is still using Spamhaus, greylisting, ix.dnsbl.manitu.net, which is not as good as it was but still VERY effective. I’ve still to go over the statistics from policyd-weight for that server.

I use content filtering in icedove, but it mostly deals with rubbish from “trusted” servers, and I get to hand train it for MY personal preferences. My own email server has policyd-weight, and so far only one known false positive in its default config.

What benefits has policyd-weight got over SA? SA can add scores for the same criteria that policyd-weight uses as well as content scores. If I wasn’t going to use SA then policyd-weight would be very interesting to me as it seems to manage DNSBLs better than my current setup. But it seems that SA can be even better than that.

Content filtering will always be error prone as content alone does not a spam email make.