Justin Mason wrote:
Daryl C. W. O'Shea writes:
Justin Mason wrote:
Maybe we should simply not publish a new active.list or rule updates after
a weekly -net mass-check -- it always seems to have significantly
different freqs, probably due to smaller amounts of mail being checked.
Hmm... that might be it. This one just caught my eye as it happened to
remove all of the new rules and a lot more other rules than normal for a
-net mass-check. I just find it odd that *all* of the new rules were
removed this time using the same submitters as the previous week.
I wonder if that indicates that old mail is being mass-checked?
Some of the corpora are very out of date -- we should probably change
the backend to be more stringent about measuring rates against only
the freshest spam...
That could be it. Theo's non-net mass-check goes back 90 days, while
his net mass-check goes back 35 days. Combined with me not getting
around to setting up net mass-checks (my non-net mass-checks go back 42
days), now that I've got a broadband connection again, we could be
introducing considerable bounce right there. This weekend's -net
mass-check based active.list should make it clear as to whether it was a
ruleqa glitch or a corpora age issue.
On the other hand, a lot of similar/identical spam seems to disappear
and reappear a few times over the course of about two months or so, so
we don't want too narrow a view.
In any case, yeah, we need to introduce some dampening in the rule
promotion/demotion to prevent rule bounce. I've got a few loose ideas
on how to do it, not sure when I'll have time to look into it though.
+1. anything you come up with will probably be better than what we
have now, ie. nothing ;)
ha -- now there's a vote of confidence if I ever saw one! :)
Daryl
