[EMAIL PROTECTED] wrote:
Of the fair amount of false negatives that get through, more than 90% of
them appear to hit on URIBL_BLACK. I have incrementally increased it
recently to a score of 5.0 (I hit on 6.0). The stuff that's still
getting through seems to be hitting on only URIBL_BLACK.
I am very tempted to bump the score of it to 6.0 or higher, as it would
drastically reduce spam, but I'd like to get any false positive feedback
on doing that first. I haven't seen any so far, but I figure others
must be doing this.
Relying on a single rule is dangerous IMHO unless you have some good
negative scoring ham rules to counteract any potential FPs. A
combination of lower scoring rules is always going to be more reliable
than a single high scoring rule.
Do you have Bayes enabled and a well trained Bayes database? I see >97%
hits on Bayes_99 and 98.5% on Bayes_80 and above. As well as
incrementing spam scores, negative scoring Bayes and AWL hits are great
for saving potential FP spam assignments when other rules hit as FPs,
especially if you've bumped the score of those rules to at or near your
spam threshold.
Another alternative is to look at adding a few custom rules to hit
against the stuff that's currently sneaking past.