[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]


On Sat, 19 Oct 2002, Craig Sanders wrote:

> On Fri, Oct 18, 2002 at 05:20:41PM +0200, Santiago Vila wrote:
> > I hope those of you who oppose so strongly to changing default
> > threshold from 5.0 will agree to downgrade it a little bit after the
> > NO_MX_FOR_FROM thing is fixed.
> no, the right thing to do is add lots of rules to spamassassin's
> local.cf so that any spam scores higher.

You are mixing different things here. I'm comparing the status before
and after the MX thing is fixed.

The right thing to do, if we do not add any rules, is to decide about
the appropriate *effective* threshold and be consistent about the choice.

> if you reduce the threshold so that nearly every message gets tagged as
> spam, you may as well just give up on having a list at all.

Yes, I agree. That's why I said "a little bit". Currently, less than 1%
of all messages I receive from the debian lists have a score between
4.0 and 5.0. This is hardly "every message".

I believe those messages should not reach the list in an automatic
fashion, since they are mostly spam. If /dev/null is not acceptable,
let's bounce them, let's moderate them. Whatever.

About the MX thing, I'll explain it again with more detail, so that you
can understand it: The current SA in murphy does not behave as it should.
It gives an extra 1.8 points to every post (for the MX thing). So the
"effective" threshold is really 3.2, since we don't see posts which
would score between 3.2 and 5.0 using a well-behaved spamassassin.

If you believe the current threshold is appropriate (and I'm beginning
to think it is, after all), you should agree, to be consistent, that
the same effective threshold is used whenever the MX thing is fixed,
which would mean reducing the real threshold to 3.2.

On the other side, if you think we should be using an effective
threshold of 5.0, then, to be consistent in your war against false
positives, you should really be asking that the current real threshold
is raised to 6.8 to compensante for the MX thing. I foresee an
unacceptable increase of spam in the lists if we did that.

> instead, spamassassin should be configured to assign a higher score to
> spam - and the only way to do that is to give it extra rules with which
> to identify spam.

Which leads us to the most recent spamassassin available, for a start.

Question: Does it compile/run under woody? (I see that the version
in testing is arch: all).

May the maintainer help the listmasters to install it in murphy?

Reply to: