I'll comment on this before anyone else does.  :-)

> Author: quinlan
> Date: Sun May 30 22:55:49 2004
> New Revision: 20683
> 
> Modified:
>    incubator/spamassassin/trunk/rules/70_testing.cf
> Log:
> removing T_SPF_PASS_NO_SBL - all of my hits are really spam

I don't think any one rule is sufficient to make a simple "does an SPF
record exist" test worthwhile.  The test needs to be paired with *actual
domain names* that are known to be good senders.

I took all SPF_PASS hits (318) in the last 14 days of corpus results and
looked at the other rules hit on those spam messages:

  318     SPF_PASS      <- spammers with SPF records
  247     HTML_MESSAGE
  239     SPF_HELO_PASS <- both!
  197     T_SPF_PASS_NO_SBL
  169     URIBL_WS_SURBL
  159     RAZOR2_CF_RANGE_51_100
  157     URIBL_SBL
  134     RAZOR2_CHECK
  130     MIME_HTML_ONLY
  121     T_SPF_HELO_PASS_NO_SBL
  121     RCVD_IN_SBL
  114     BAYES_99
  112     T_RCVD_IN_SBL
  106     URIBL_BE_SURBL
  103     T_RATWARE_RCVD_PF_1
  98      CLICK_BELOW
  ... long tail

I don't think any small set of rules is sufficient.  And if you include
too many rules, then the entire point of having a negative rule is
missed.  We should be attempting to couple SPF pass with specific names.
For example, it should be required for our default whitelist.

Daniel

-- 
Daniel Quinlan
http://www.pathname.com/~quinlan/

Reply via email to