Sn!per wrote:
Hi all,
Am using SA version 3.2.4 . So far working great.
I am comparing the headers on two of my mails. The first mail is NOT a spam,
and X-Spam-Status is like so:
No, score=2.7 required=10.0 tests=RCVD_NUMERIC_HELO,RDNS_NONE autolearn=no
version=3.2.4
The second email is a spam, and its X-Spam-Status looks like this:
Yes, score=24.4 required=10.0 tests=DCC_CHECK,DIGEST_MULTIPLE,
PYZOR_CHECK,RAZOR2_CF_RANGE_51_100,RAZOR2_CF_RANGE_E8_51_100,RAZOR2_CHECK,
RCVD_IN_BL_SPAMCOP_NET,RCVD_IN_PBL,RCVD_IN_SORBS_DUL,RCVD_IN_XBL,RDNS_NONE,
URIBL_BLACK,URIBL_JP_SURBL,URIBL_RHS_DOB,URIBL_SC_SURBL,URIBL_WS_SURBL
autolearn=spam version=3.2.4
Question: Why is it that on the header of the ham, autolearn=no ?
Because with a score of +2.7 it didn't meet the score criteria for
either spam or nonspam learning. SA only autolearns low scoring nonspam
and high scoring spam. Everything "in between" it figures might be
miscategorized, so it doesn't autolearn it.
By default, the "learning score" needs to be under 0.1 to autolearn as
nonspam. It needs to be above 12 to learn as spam (with at least 3
header-rule points and 3 body rule points).
See also:
http://spamassassin.apache.org/full/3.2.x/doc/Mail_SpamAssassin_Plugin_AutoLearnThreshold.html
Note: the "learning score" is the score the message would have gotten if:
1) bayes was disabled (includes scoreset change)
2) the awl doesn't count.
3) any rules with tflags learn or userconf (ie: white/blacklist
rules) don't count.
SA uses that score for deciding to autolearn or not to prevent
self-feedback on the learning systems, and to prevent a mistaken
whitelist_from from causing a lot of spam to autolearn as nonspam.
