On 07/21, Michael Scheidell wrote: > but, this rule seems to be way too aggressive. > > 50_scores.cf:score FS_REPLICA 1.630 3.599 2.028 3.599 # n=2 > 50_scores.cf:score FS_REPLICAWATCH 3.237 1.715 1.733 3.015 # n=2
> 72_active.cf:header FS_REPLICA Subject =~ /replica/i > 72_active.cf:describe FS_REPLICAWATCH Subject says Replica watch Scores are automatically generated based on optimal performance based on actual emails: http://wiki.apache.org/spamassassin/NightlyMassCheck Statistics of the results are regularly generated: http://ruleqa.spamassassin.org/?daterev=20110716&rule=%2FFS_REPLICA SPAM% HAM% S/O RANK SCORE NAME 7.5986 0 1.000 0.92 3.24 FS_REPLICAWATCH 9.8513 0.0143 0.999 0.92 1.63 FS_REPLICA So FS_REPLICA hits 9.8513% of spam while hitting 0.0143% of ham. That's less than 1 in 2500 hams (0.04%), which is the overall false positive rate SpamAssassin aims for. But yes, it does look like this pair of rules could be improved. It is an unfortunate problem that score generation doesn't have a good way of accounting for heavy overlap. > anyone else think that ANY rule that scores above a 3 is asking for trouble? There has been talk of limiting that kind of thing. -- "Where are you going and what do you wish?" - The Old Moon, to Winkin' Blinkin' and Nod http://www.ChaosReigns.com
