Specs:
debian 8
spamassassin 3.4.1
MailScanner 5.0.3
We have some mailservers which (auto and manually) learn via sa-learn from different hosting-servers.
Spam will mostly get tagged and moved to the spam folder as wanted.
SA-learn spam will be triggered for emails in the spamfolder. SA-learn ham will be triggered for emails which the user replied to.
Now we have problems that some mails are getting tagged as spam, which should NOT get tagged as spam (no-reply@steampowered, no-reply@amazon and so on). Mostly becouse of BAYES 999 which includes BAYES 99. We think becouse some users move those mails into the spam folders, which we will learn as spam for global spam / ham decissions.
We would like to learn ham from those "no-reply mails", but we have no "good allround conclusion" on "which" mails we can learn.
Is there someone here, who have some thoughts or maybe a good solution for our problem?
We thinked about learning ham from mails which where moved from spam to inbox, but this would only help for the wrong spam learned mails before.
We also look for a trusted whitelist source (does anybody knows one?) which we could additionally include into the spamassasin / mailscanner to get better spam / ham results.
We appreciate any help
kindly regards