On 6 Apr 2021, at 11:48, Steve Dondley wrote:
I have emails that have been flagged as spam in the past but that are
still getting through, presumably because the servers are on some
DNSWL.
Example:
X-Spam-Status: No, score=0.9 required=5.0 tests=BAYES_99,BAYES_999,
DATE_IN_PAST_03_06,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,
HTML_IMAGE_RATIO_02,HTML_MESSAGE,RCVD_IN_DNSWL_HI,RCVD_IN_MSPIKE_H2,
SPF_HELO_NONE,SPF_SOFTFAIL shortcircuit=no autolearn=no
autolearn_force=no version=3.4.2
What's the recommended way to handle these? Do I turn on shortcircuit?
Certainly NOT for Bayes, since it is fairly common for Bayes to be
simply wrong on essentially random messages. I personally don't think
shortcircuit qualifies as a generally useful feature in the modern
world.
Do I bump up the score for BAYES_99, BAYES_999? Or might there be a
way to ignore DNSWL scores if they have a high bayes score?
Rather than constructing a meta-rule for just that case, I adjust
scores:
score RCVD_IN_DNSWL_HI -2
score BAYES_00 -1.0
score BAYES_999 0.5
Because DNSWL has problematic sources, BAYES_00 is not really a great
ham indicator, and BAYES_999 is better than the default score gives it
credit for.
YMMV!
--
Bill Cole
b...@scconsult.com or billc...@apache.org
(AKA @grumpybozo and many *@billmail.scconsult.com addresses)
Not Currently Available For Hire
