mail scored 6.5 points lead to autolearn=spam

["Philipp Snizek, seaan.net ag" <ps...@seaan.net>]

Hi

X-Spam-Status: Yes, score=6.5 required=3.0
tests=BAYES_50,INVALID_DATE,
	
MSGID_FROM_MTA_ID,NO_REAL_NAME,RCVD_IN_NJABL_DUL,RCVD_IN_SORBS,
	RCVD_IN_SORBS_DUL,TW_KT,URIBL_AB_SURBL,URIBL_SBL
autolearn=spam 
	version=3.0.1

I thought the autolearn=spam is at a score of 12. 
What are the reasons that it still is autolearned?

thank you
Philipp

Re: mail scored 6.5 points lead to autolearn=spam

[Matt Kettler <mk...@comcast.net>]

At 03:05 AM 2/17/2005, Philipp Snizek, seaan.net ag wrote:
>X-Spam-Status: Yes, score=6.5 required=3.0
>tests=BAYES_50,INVALID_DATE,
>
>MSGID_FROM_MTA_ID,NO_REAL_NAME,RCVD_IN_NJABL_DUL,RCVD_IN_SORBS,
>         RCVD_IN_SORBS_DUL,TW_KT,URIBL_AB_SURBL,URIBL_SBL
>autolearn=spam
>         version=3.0.1
>
>I thought the autolearn=spam is at a score of 12.
>What are the reasons that it still is autolearned?

RCVD_IN_SORBS -- hmm, that's an odd hit.. There's no rule named that in SA 
3 that I can see... only __RCVD_IN_SORBS which should be a no-show.. Can 
you check your rule files and see what's up? Any chance you've got the 2.6x 
rules with the 3.0x code?


I can point out that looking at the message score is almost completely 
irrelevant to the autolearner's behavior...

Remember, SA does not autolearn based on the message score.. it calculates 
the score the message would have gotten if bayes were disabled. Disabling 
bayes doesn't just eliminate the BAYES_50, it also changes the score of 
every rule involved.

Thus you need to re-calculate the score using score set 1 (the four scores 
in the score line are set0, set1, set2, and set3. In terms of bayes and 
network checks 0=neither, 1=net only, 2=bayes only, 3=both.)

Switching to a different score set can cause *huge* differences.. I've seen 
10 point differences.

Since I'm concerned your rules are corrupted, I don't know what your scores 
are.. clearly something's not right as RCVD_IN_SORBS doesn't exist. Adding 
up the other rules I get a score of 5.532 for set3, clearly this is not the 
message score you got...

I can add up the set1 scores and get 6.622, but that's not useful as I 
don't know what RCVD_IN_SORBS is doing matching, and the set3 score isn't 
correct. Clearly you've got scores that don't match the ones I'm working with.