[nylug-talk] SpamAssassin best practices

[Ron Guerin]

Raj Goel wrote:

>    I've had SA running for a while - using Razor & sa-updates.
> 
>   I'm seeing a number of spam get thru SA and no amoumt of sa-learn --spam /var/spool/spam seems to do the trick.

Agreed, I think Bayes, or at least the implementation in SA has outlived
its usefulness.  It seems to do more harm than good now. (ie: it's not
doing any good, and Bayes poisoning is causing it to do harm)

>    How are you handling the current rise in spam?

I'm not aware of a "current rise", only a continuing one.  That may only
be because my current configuration is dealing with it well.

To Razor, add Pyzor, and DCC.  DIGEST_MULTIPLE is a wonderful thing.
You might try adding some additional blocklists with very low scoring,
and for image spam, FuzzyOCR.  I've had to tweak the wordlist a little,
but have had mostly good results with it.

- Ron