forgot to mention - you *never ever* will be able to reach a compareable
hitrate with your local scanning compared to a global bayes feeded with
currently 85000 spam and ham samples and we are *really* a small player
in filtering
not only if it comes to filter out spam - also to prevent false
positives based on ham samples
/etc/mail/spamassassin/local-*.cf
score BAYES_00 -3.5
score BAYES_05 -2.0
score BAYES_20 -1.0
score BAYES_40 -0.5
score BAYES_50 1.5
score BAYES_60 3.5
score BAYES_80 5.5
score BAYES_95 6.5
score BAYES_99 7.5
score BAYES_999 0.4
0 62431 SPAM
0 21895 HAM
0 2585854 TOKEN
BAYES_00 14394 74.24 %
BAYES_05 415 2.14 %
BAYES_20 490 2.52 %
BAYES_40 474 2.44 %
BAYES_50 1495 7.71 %
BAYES_60 195 1.00 % 8.74 % (OF TOTAL BLOCKED)
BAYES_80 158 0.81 % 7.08 % (OF TOTAL BLOCKED)
BAYES_95 147 0.75 % 6.58 % (OF TOTAL BLOCKED)
BAYES_99 1619 8.35 % 72.56 % (OF TOTAL BLOCKED)
BAYES_999 1442 7.43 % 64.63 % (OF TOTAL BLOCKED)
DELIVERED 26962 94.28 %
DNSWL 26683 93.31 %
SPF 18807 65.77 %
SPF/DKIM WL 8088 28.28 %
SHORTCIRCUIT 9185 32.12 %
BLOCKED 2231 7.80 %
SPAMMY 2119 7.41 % 94.97 % (OF TOTAL BLOCKED)
Am 14.03.2016 um 19:40 schrieb Reindl Harald:
Am 14.03.2016 um 18:45 schrieb P. Gueckel:
> Patrick O'Callaghan wrote:
>> How do you expect them to filter spam without
> scanning your messages?
>
> Good point. Scanning is a privacy breach that likely
> violates privacy legislation. All the more reason to
> disable the feature and do your own scanning
*lol*
we have days where 500000 mails for a few hundret users are rejected
straight ahead by postscreen-scoring on the MTA long before
content-filters become a topic
have fun receive that all, scan it local and what about false positives
you overlook - on our MX servers *any* message above 8.0 socre points is
rejected unconditionally, so a sane sender generates *it's own bounce*
to his enduser and nothing is silently dropped
between 5.5 and 7.9 points it get tagged *but not* moved somwehere, just
a SA header and a mark in the subject - the user is responsible at it#s
own how to handle such mails by headers and/or subject
8-10% of all mails not rejected straight away because bad reputation
based ona mix of 50 combined dnsbl/ndswl are rejcted by milter
have fun doing that local where you *must not* reject and backscatter