Anne wrote: > Hi, > > DATE_IN_PAST_48_96 was taken out since 3.2.x. Why?? > What happens with spam between 48 and 96 hours in the past? Looks like it was dropped due to its horribly poor performance. I can't confirm why it was dropped, but I can point to strong evidence the rule was worthless.
In the 3.1.x set0 mass-checks it had a S/O of 0.649, which isn't significantly different from the whole set's S/O of 0.700. In essence, the rule seemed to match spam and nospam with more-or-less equal probability. To the extent it differed from the distribution of the test data, it favored matching nonspam. (ie: the S/O of the rule is less than the S/O of the test data) Sidenote: S/O is the Spam/overall hit ratio. If you multiply by 100, you've got what percentage of the email the rule matched was actually spam.