Justin Mason wrote:
James Gray writes:
On Wed, 26 Mar 2008 12:09:47 pm D Hill wrote:
Now your confusing the subject. The previous response you made was from:
From: James Gray <[EMAIL PROTECTED]>
Now you are using:
From: James Gray <[EMAIL PROTECTED]>
BOTH of those domains point to an MX that has a CNAME to:
smtp.mas.viperplatform.net.au
So I use one mail client and occasionally forget to set the correct profile to
send as.. sorry.
No. None of the domains have any references to smtp.mas... as an MX record.
They all point to mail.mas... and they all have a default TTL of 38400 for
the zone.
Besides, this still doesn't change my original point that SORBS have a habit
of listing addresses (justifiably or not) then attempt to extract money to
remove the listing. That's just extortion, not a good RBL.
It's worth noting that SpamAssassin has *never* used the SORBS sublist
that requires this payment. We do not endorse that "pay-to-remove"
concept.
Also, folks -- regardless of how RFC-anal his DNS records are, that
doesn't change the fact that SORBS DUL is listing his IP space
incorrectly. This is *definitely* a false positive for SORBS. So please
stop rattling on about that end of things.
How do you know that sorbs is listing his IP space incorrectly? he
didn't show what IP space he was talking about.
Anyway -- the SORBS DUL sublist is still in SpamAssassin due to its
success at hitting spam with few false positives. Compare their accuracy
ratings against the nearest analogue, RCVD_IN_PBL:
http://ruleqa.spamassassin.org/20080322-r639965-n/RCVD_IN_SORBS_DUL/detail
http://ruleqa.spamassassin.org/20080322-r639965-n/RCVD_IN_PBL/detail
SPAM% HAM%
36.2977 488419 of 1345591 messages 0.1094 87 of 79523 messages
0.997 0.91 0.00 RCVD_IN_SORBS_DUL
66.7720 898478 of 1345591 messages 0.1823 145 of 79523 messages
0.997 0.87 0.00 RCVD_IN_PBL
that's a 99.7% accuracy rating, and generally hitting on the low-scoring
spam too, which is the most valuable. Having said that, it has a very
low score; in set3 it provides only 0.91 points. That's the GA
compensating for its false positives.
Nowadays it's mostly subsumed into PBL:
overlap spam: 91% of RCVD_IN_SORBS_DUL hits also hit RCVD_IN_PBL;
49% of RCVD_IN_PBL hits also hit RCVD_IN_SORBS_DUL
so it's likely that the PBL would compensate entirely for the loss
of SORBS DUL, if we were to remove it. But we don't know.
If you like, open a bug on our bugzilla suggesting that for the next
rescoring run (for 3.3.0), we disable SORBS_DUL and see if accuracy
survives ok, and we will try that out.
--j.
