[Mimedefang] SpamAssassin 2.43

[Nels Lindquist]

On 23 Nov 2002 at 15:12, David F. Skoll wrote:

> Has anyone noticed that the scores in SA 2.43 seem a bit conservative?
> I'm seeing more spam slipping through than with previous versions.

Their stated goal with the 2.4x release was to try and reduce false 
positives for mailing lists, subscribed newsletters and the like.  In 
order to do that, they had to reduce the magnitude of some "spammy" 
characteristics so that the compensation scores have a chance to 
work.

I actually did a fair amount of analysis on my own corpus, and it 
turns out that the raw number of messages scored at any given value 
didn't change much (if at all) but that certain specific messages did 
change in value quite dramatically.  On balance, more false negatives 
were pushed in to the spam category (score >=5) than vice versa.

You're not using any of the RBL tests, are you David?  I find the 
Razor checks to be unacceptably slow, but the RBL tests don't slow 
things down nearly as much, and they definitely help.

----
Nels Lindquist <*>
Information Systems Manager
Morningstar Air Express Inc.