Since last update “train good” always fails. As soon as it has trained, the mail comes back as unread and gets filtered as spam. Nothing suggested in the help is working.
I have to quit SpamSieve then it starts behaving normally again for a while, but as soon as these spam-mails that really looks like spam but aren’t comes in - things break down again.
Did you see this help page? The most common reason this happens is that another filter (or perhaps SpamSieve running on another Mac) is filtering the message as spam. The Mac where you did the training shouldn’t be filtering the same message again, so even if Train as Good weren’t working it wouldn’t cause the result that you describe. You can see in the Log window whether the training took place and whether or not the Mac predicted the same message to be spam again after the training.
Yes, “Nothing suggested in the help is working.”
Also when marking a mail as good, SpamSieve deleted a mail in the process. It looks like it removes mail from spam-box first, then trains, then puts it back as a new unread mail. The putting back thing never happened.
That is really concerning to me. I have now updated my Mac to 15.2 and will try again and see what happens (spamSieve was always on the latest non-beta).
This doesn’t happen every time so very hard to reproduce.
Will also double-check the log window.
Are you referring to a SpamSieve update or a macOS update? There were no changes to the Train as Good command in SpamSieve 3.1.
Your post didn’t say which help page you read or anything about what you did. Are there multiple Macs/PCs that access this mail account?
That’s not what it does. It trains the message and then asks Mail to move it to the inbox. As noted here, there is a Mail bug where sometimes it deletes messages instead of moving them if the Junk mailbox is not set to be stored on the server. Did you mean to say that the message was removed from the Junk mailbox or was it actually in a “spam-box”?
It’s also possible that the message did get moved, but to the inbox of another account or of a disabled account. The most common reason messages disappear after training as good is if another device sees the message in the inbox as a new spam message and moves it to a local Junk mailbox on that device; the other devices see this as a deletion because the message is removed from the server when it’s moved.
SpamSieve stores a backup of the message data when you train a message, so you can restore the message to Mail.
The SpamSieve update.
Thanks for clarifying this. It wasn’t super-important email, but good to know there are backups.
I still don’t know why it is misbehaving or how I can fix it. As long as it filters the bad ones as spam I can live with this situation, but of-course would love if it could learn when it makes mistakes.
Yes - I did see that help page and nothing in the help (that help page describing the problem I’m having) is working. The page is this one: SpamSieve Manual: Why does a message disappear or go back to the Junk mailbox after I train it as good?
Yes, many macs/iOS clients access the mail - one has spamSieve, the others do not move anything (has any filters or similar applied).
Was it only the one message that disappeared? The others keep cycling between Inbox and Junk?
So far we have not definitively established that it made a mistake or that it didn’t learn. That’s why I suggested looking at the Log window and searching for one of these messages. Does it say that the message was Predicted: Spam before you trained it as good? Are there any Predicted log entries for that message after you trained it as good?
Did you do all the things it suggests? One approach to troubleshooting this would be for you to go through that page and explain to me (e.g. with screenshots) each of the items and why you think it doesn’t apply to your situation.
Or you could use the Save Diagnostic Report command in the Help menu and send me the report file, as described here, and maybe I will be able to see what’s happening or see that you didn’t complete one of the steps on that page.
Does the problem still occur if you turn them all off before training the message as good?
Yes, only one disappeared, the others where cycling.
Yes, I totally forgot! It does in fact say it predicted it to be green (even before I trained it as non-spam). But then how could it end up in the spam-folder?
Perhaps you have been correct all along and there, in fact, is another client accessing the mail and doing filtering. 
Will try to turn everything off and see if I can recreate the problem and be 100% sure it really is SpamSieve that causing this.
Sorry for wasting your time but thanks for the support!