On 17 Feb 2003, Nix wrote:
> On Mon, 17 Feb 2003, Theo Van Dinter muttered drunkenly:
> > score RCVD_IN_RFCI 0 1.771 0 1.249
> > score RCVD_IN_ORBS 0 0.458 0 0.121
>
> Now those I believe; lower after Bayes kicks in...
>
> > score RCVD_IN_DSBL 0 2.225 0 4.295
>
> ... but this is rather hard to believe. The DSBL becomes an almost
> certain spam-sign only once Bayes is active?
>
> What's going on here?

Well 4.2 is pretty close to certain. On the other hand a quick check shows
false positives from these two entries:

http://dsbl.org/listing?ip=219.88.241.226
http://dsbl.org/listing?ip=202.42.247.99

and I saw another one as well :)

>From what I can tell the DNSBLs are being checked at the same time as the
mass-check rather than when the email is being received.

In the meantime (even a week or two) the ips could have been removed from
the DNSBL . Alternativly the entry could have only been recently added but
the false positive hit could be exagerated since all the corpus emails
with that ip would hit. For something like bl.spamcop.net this would be
especially distorting.

I thought Bayes didn't learn off the RCVD_IN_DSBL entry since it isnores
the headers spamassassin puts in.

-- 
Simon Lyall.                |  Newsmaster  | Work: [EMAIL PROTECTED]
Senior Network/System Admin |  Postmaster  | Home: [EMAIL PROTECTED]
Ihug Ltd, Auckland, NZ      | Asst Doorman | Web: http://www.darkmere.gen.nz



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk

Reply via email to