Training issues

I’m a longtime SpamSieve user. I just installed it on my new M1 MacBook Pro. I seem to be having an issue with the SpamSieve rules and false positives.

Maybe it’s simply a matter of building up the corpus on my new machine, but there are now hundreds of older messages (going back years) in my local Apple Mail spam folder. I’ve gone through them and trained them as Good, but an hour or two later they reappeared in the Spam folder along with more older messages.

ADDENDUM: I just moved them all out and they’re all coming back in again!

Another issue is that when I move them back using the “Train as Good” script, they all get marked as unread, and I have to go through and mark them as read again.

Something else must be going on because SpamSieve doesn’t go through old messages unless you tell it to. First, I would suggest looking in SpamSieve’s log to see whether it predicted those messages to be spam. If not, the problem may be with another app or device or server filter. This page may also be helpful.

This also makes it sound like another Mac/device is moving the messages because Mail only auto-applies SpamSieve to new messages in the inbox.

There’s a setting to control this, but you should only be training them as good if SpamSieve on this Mac predicted them to be spam.

You’re right. Nothing got flagged by SpamSieve as actual Spam. There’s nothing in the log but “Predicted: Good”
The Rule I created for SpamSieve is local to my machine and my Spam folder is local, so I don’t see a conflict on another device that could be causing this.
At this point it looks like most of the messages that are still in the Spam folder (121 at last count) lacks body. They’re just the header information. And about 90% of them are from my work (G Suite) account.

I stand corrected… There were quite a few that got flagged as Spam. But since I reduced the number in the Spam folder, I decided to prune the old messages down and deleted the ones that remained in Spam. I’ll see what the behaviour is with new messages.

Right, unless there’s another Mail rule that uses that mailbox. Messages can bypass processing by SpamSieve if the plug-in is not enabled in Mail’s preferences. Perhaps there was a window in time during setup when that was the case?

That doesn’t explain why the messages would go back to the Spam mailbox after training, though, unless you had manually re-applied the rules. One possibility is that you’re running into one of the macOS 10.15/11.x Mail bugs with moving messages. It could be that the message isn’t moving back to Spam but rather that it’s not properly moving to the inbox in the first place. This would also match up with what you reported about the non-header information being lost. I recommend that you enable SpamSieve’s backup feature to protect against this, and you may also be able to avoid the Mail bug by changing to use a spam mailbox that’s on the same server as your inbox.

If you would like me to look into why the messages were classified as spam, please send in a diagnostic report.