I had been running SpamSeive 3.0.2 on MacOS 13 (and 12, 11, 10…) with no issues. Immediately upon first use of Mail after the upgrade of the OS, SpamSeive is being excessively aggressive in marking things as spam.
After Filtering, SS left 1 (one!) email in my Inbox. Looking through all my Junk folders (one per mail account), I have well over 100 emails that before the upgrade would not have been Filtered (i.e., they are NOT Junk).
Why is SS being so overly aggressive?
edit: speling
Maybe it wasn’t. Does the Log window say that SpamSieve predicted these messages to be spam?
Please make sure that you’ve turned off the SpamSieve rule, as described here, as on macOS 14 it will move everything to Junk without consulting SpamSieve. SpamSieve will usually turn it off for you, and explain why, but a lot of people seem to turn it back on, thinking they know better.
Not clear exactly how to read this log. But right off the bat after the upgrade, we see that rather than using training data, it has deigned to use some ‘blocklist’ to tank the first thing it looked at - an email that may be spam to some people, but is not to me:
This is saying that SpamSieve classified the message as spam because you had previously trained a message from Choice Privileges as spam.
If you think messages may have been trained as spam in error, you can review the Corpus window to make sure that the messages are categorized correctly.
You say that I did, but I believe that I did NOT. Certainly, within the last week, I had received at least one email message from Choice Privileges that SS did not tank.
Perhaps SS mistakenly created a batch of training data in my stead?
Since my initial report earlier today, SS CONTINUES to classify non-spam as spam. Including such important entities as American Express, Samsung, and even Google Voice. I MOST CERTAINLY 100% NEVER trained SS to treat these as spam.
We need a new operating theory as to why this is occurring, and we need a way to fix it.
You can search the Log window for Choice Privileges, and it will show you when and how the blocklist rule was created (unless it was created before SpamSieve 3 was installed—in that case, check the Blocklist window).
SpamSieve will auto-train the corpus, but it does not auto-train blocklist rules because that would be dangerous.
What does the log show for those messages?
Again I do not believe your assertion to be correct. Searching throughout the log for ‘choice’ turns up three entries:
This is the same log entry from your earlier screenshot. It was predicted by the blocklist.
It’s telling you that the rule was not created with SpamSieve 3 (the version that introduced the Log window). So the rule must have been created earlier. As I said above, you can search for it in the Blocklist window, and double-click the rule, and it will say when it was created. There are also text log files for SpamSieve 2 in the folder:
/Users/<username>/Library/Logs/SpamSieve
which would show which message was trained as spam that caused the creation of this rule.
That seems to be a coincidence because the rule was created before SpamSieve 3, and you said you were already using SpamSieve 3.0.2 with macOS 13, 12, 11, and 10. (So many macOS upgrades since SpamSieve 3.0.2 was released in November?)
Sorry - that OS thing was a misunderstanding. I was using SS 3.0.2 previous to upgrading to MacOS 14. I have no idea what versions of SS I was using earlier than MacOS 13.
And again, NO. Previous to upgrading MacOS 3.0.2, I was routinely getting emails from Choice Privileges without them being flagged as Spam.
OK, from the blocklist window:
So I guess as you say, the rule had been previously created. Yet I was routinely getting unblocked emails from them. So if a rule was created, then marked as good (some time in the distant past), then this morning changed to marked as spam, then again marked as good, how would that look in the Blocklist or the Log?
And this one:
It fits what you say… the rule was created evidently on 6/19/21. Yet I was routinely getting emails from them without being moved to the Spam file. So perhaps the rule was retrained as Good on some date between 6/19/21 and today? (perhaps on 6/20/21) Note all these that I have, that I have not retrieved from the associated Junk folders - they rather just appeared in my Inbox (other than the one from this morning, which was shuttled into Junk):
edit:speling
OK, how about this one - another sender that I was routinely getting emails, without being shuttled to the Junk folder:
I guess you’d think that it was in the blocklist, but it ain’t:
You can look back in the log to see for sure why these other messages were Predicted: Good, but my two guesses are:
If you had trained one of these messages as good at any time after the blocklist rule was created, SpamSieve would have disabled the blocklist rule. Since the rule was enabled (at the time the message was Predicted: Spam) we know that’s not the case (unless you had manually re-enabled the rule, which I’m assuming you didn’t).
Again, it looks like the rule was enabled as of 12/4, so that means you hadn’t trained any messages matching that rule as good after 6/19/21. The complete history of the rule is in the log files that I mentioned, if you want to search for it.
This one has a score of 96 (vs. 99) so that means it was caught as spam by the Bayesian classifier, not the blocklist. You can click on the log entry to see which words in the message looked spammy to SpamSieve.
The other thing to investigate is that, if you were previously getting messages from this sender, SpamSieve had probably auto-trained allowlist rules for the name and address. So I wonder whether those were different with this particular message or whether you trained something as spam that caused the allowlist rules to be disabled.
By the way, you can check the Locked box next to an allowlist rule if you want SpamSieve to keep it enabled even if you train a message spam message that matches the rule.
All right… this is getting ridiculous. I get that you don’t think there is anything wrong with SS’s behavior here. I say you are in denial. Hundreds of emails that would have ended up in my inbox right up until I upgraded to Sonoma are now shuttled to the Junk folder. And this just goes on and on for nearly every sender. This is a drastic overnight change in behavior.
I am now tied to doing email only on this single computer, because I would otherwise miss all the emails that this one is Junking.
How to I tell SS to throw out its current training database, and start fresh?
So far you’ve mentioned three specific examples of good messages going to Junk. From the log, it looks like all of them were from senders that you had previously trained as spam. Now, SpamSieve doesn’t only rely on the blocklist, and with training it can learn to distinguish between good and spam messages from the same sender, but it’s not surprising that if you train them as spam they will go to Junk.
It’s not clear why you are seeing a sudden change when updating to macOS 14 when you didn’t even change the version of SpamSieve that you were using. The only thing I can think of is that maybe these senders were in your Contacts and that something happened to your Contacts database (or SpamSieve’s access to it) when you updated macOS. If you use the Save Diagnostic Report command in the Help menu and send me the report file, as described here, I can look into what might be going on here.
You can follow the Resetting SpamSieve instructions at the bottom of this page. However, I don’t recommend doing this just yet, because the diagnostic report may reveal some other issue, and I don’t want you to spend time retraining when the problem might be something else.
