Bogofilter accuracy plummets starting around March 10, 2010

[David Relson]

On Tue, 13 Apr 2010 04:23:47 +0400
Dmitry wrote:

> David Relson wrote:
>  > "Invisible headers" is not a term I recognize.
> 
> Headers like "To, cc, from, subject, date" -- are visible in almost
> all MUA. Everything else is usually invisible. When you allow
> bogofilter to process any invisible headers you pollute the database
> with random data and make spam detection innacurate. Real spammer has
> 2 way to break spam filter: 1. To make headers the same way as known
> mail user agents do; or 2) to make random headers in each message.
> So, in any case it makes more harm than good from the viewpoint of
> spam detection.
> 
> 1) You allow random data
> 
> > pipe your message through an appropriate "egrep -v
> > "^(this|or|that):" command.
> 
> To send millions messages through pipe with egrep? No, thanks. It is 
> unnecessary load on the server.

Dmitry,

Bogofilter recognizes a number of headers (to, from, return-path, etc)
and creates special symbols.  This gives them special significance.

Other header tokens don't get the same treatment, hence are treated
similarly to tokens in the body.  

I don't perceive that bogofilter has a problem.  It does well for the
thousands of spam arrriving daily at my domain.

Adding millions of header tokens is no different from adding millions
of body tokens, as far as I can tell. 

If you want bogofilter to provide special handling for other special
tokens, the source code is available and you can customize to your
heart's content.

Regards,

David