Current situation: To avoid false positives -- and to avoid inadvertently deleting good emails -- I have Bayesian set to be relatively lenient in evaluating messages. Many spams are marked as "good," but good messages are rarely marked as "spam." That's important for me, as my good messages often include words that occur frequently in spam.
I want to be able to quickly evaluate the "good" messages to sort through which are really spam. Currently, I sort by "status," then sort by "learning." That makes "learning" the primary sort criteria and "status" the secondary criteria, which puts all the ones that are most questionable together.
What should happen is that all things with the same "learning" rating are grouped together, and within each of those groups, the messages are sorted by the "status" column. I then should be able to glance through the spam that is misidentified as "good," find the real good messages, either mark them as "Friends" or create/modify a filter so they will be correctly identified in the future, repeat the sorting, select all the spam at once, and then change its learning column to "spam" before clicking "process mail."
The problem is that when the status is "Friends," it appears that once the mail is resorted by "learning," the "Friends" designation in the "status" column isn't being taken into account in the sort. The "status" that appears to count is what those messages *would* have had if they had not been on the Friends list. So messages that are on the Friends list -- and known to be good -- are interspersed among all the spam that was misidentified. I can't select the whole list and change the learning rating. I have to go through and select small groups of spam, then skip over some Friends, then select another group of spams, etc.
When the "status" column is the primary sorting criteria, "Friends" are all grouped together. I would like that to be the case when it is the secondary criteria, too.
