AW: RP_MATCHES_RCVD badly scored

[Support SpamAssassin <Sp...@jam-software.com>]

@Kevin:
Thanks for your answer.
We will watch this rule further more and may discuss it on the users list.
As a quick solution we created our own channel for us and our users in which we will re-scored the rule.

Best regards,

Harald

-----Ursprüngliche Nachricht-----
Von: Kevin A. McGrail [mailto:KMcGrail@PCCC.com]
Gesendet: Montag, 22. Juni 2015 18:29
An: ruleqa@spamassassin.apache.org
Betreff: Re: RP_MATCHES_RCVD badly scored

On 6/22/2015 10:12 AM, Support SpamAssassin wrote:
> Hi,
> I'm Harald Binkle, the project leader of SpamAssassin for Window at Jam Software GmbH.
> My personal contact to the SpamAssassin community is long ago.
> Daniel Lemke did further SpamAssassin development during the last few years for us.
>
> However, I'm back and I'd like to get involved again if you don't mind.
>
> Recently (approximately during the last two weeks) we noticed that more spam came through than before.
> We were able to track to problem down to a single rule:
> RP_MATCHES_RCVD This rule applies to a lot spam messages. I think giving a score of -1.4 for such a single rule is very much.
>
> Is there a way to find out which score that rule had in the past?
>
> How can I submit a suggestion to lower the score of that rule because it's really misleading with -1.4 ?
> (I would give a score like -0.2 to that rule.)
>
Hi Harry,

A couple of points below to help:

The first place to start would be ruleqa.spamassassin.org and check the
S/O.   The rule has an S/O of 0.01 hitting on about 25% of the spam and
.25% of ham in the masscheck corpora.  That's a pretty good S/O for a
Ham rule and currently it is scoring -1.05.   See
http://ruleqa.spamassassin.org/20150621-r1686693-n/RP_MATCHES_RCVD/detail

Second, if you want, I have rule files going back to 2007 at least on sa-update.pccc.com and I can send you the list of files.  You can then download, untar and compare the scores for the rule to your heart's content.

Third, you might get more input if you posted about this on users@.

Fourth, assuming you know you can override the score in your own local cf files.

Fifth, that rule is auto scored from the genetic algorithms employed in masscheck.  There are two avenues to improve that: 1) join the masscheckers with your corpora so as to lower the scores.  I must admit I'm behind in approving and setting up masscheckers though. 2) we can argue this on dev or users and see if a ceiling on the rule is appropriate.

Regards,
KAM



________________________________



--------------------------------------------------------
JAM Software GmbH
Managing Director: Joachim Marder
Am Wissenschaftspark 26 * 54296 Trier * Germany
Phone: +49 (0)651-145 653 -0 * Fax: +49 (0)651-145 653 -29
Commercial register number HRB 4920 (AG Wittlich) http://www.jam-software.com

________________________________



----------------------------------------------------
JAM Software GmbH
Geschäftsführer: Joachim Marder
Am Wissenschaftspark 26 * 54296 Trier * Germany
Tel: 0651-145 653 -0 * Fax: 0651-145 653 -29
Handelsregister Nr. HRB 4920 (AG Wittlich) http://www.jam-software.de

Re: AW: RP_MATCHES_RCVD badly scored

["Kevin A. McGrail" <KM...@PCCC.com>]

On 7/9/2015 4:50 AM, Support SpamAssassin wrote:
> @Kevin:
> Thanks for your answer.
> We will watch this rule further more and may discuss it on the users list.
> As a quick solution we created our own channel for us and our users in which we will re-scored the rule.
An easy solution!