I’ve been using SpamSieve since 2008, and now have a drone setup running SpamSieve 2.9.48 in Mail.app 16.0 under macOS Monterey (at my desk computer), with an iPhone SE (2020) and an older MacBook Air (macOS Catalina) for running around and travel. Mine is an IMAP account from a local ISP (cruzio.com here in Santa Cruz), which as far as I can see has little or no server-side filtering.
This setup seems to work just fine: there are very few false negatives or positives generated. I correct errors, and continue to be appreciative of your work.
But I wonder: might my system’s performance still be improvable, or be less effective than might be expected with this setup? Asked differently: is this a reasonable level of performance? I don’t see a way to show the Statistics panel in this composition box, but the Statistics panel contains these figures:
Filtered Mail:
4,386 Good Messages
2,278 Spam Messages (34%)
12 Spam Messages Per day
It’s common to get 99.9% filtering accuracy, but 99.6% is not bad. Given about 25 mistakes total this year, it may not be worth spending the time to try to reduce the error rate. However, if you’d like to do that, I would suggesting going through this page or using the Save Diagnostic Report command in the Help menu and sending me the report file, as described here so that I can see whether SpamSieve should be doing something differently or whether there’s something that should be adjusted in your setup.
Thanks for the report. Some of the spams got through because they used names and addresses that had previously only sent you good messages. SpamSieve errors on the side of safety, and I think this is what most people want, so I would not recommend changing anything there.
For some of the spam messages, I see areas where SpamSieve could improve its analysis, and I’ll be making some changes in a future version.
Some of the messages (particularly the false positives) were “bad luck” or cases where the results might have been better with more recent data. It might help to reset the corpus and retrain with a smaller number of recent messages. I’m not sure that’s worth it, though, and first I would try setting the spam catching strategy back to the middle—it’s set to be extra aggressive right now.
I also see a bunch of cases where you trained messages as good that SpamSieve had already predicted to be good. I’m not sure why that happened, but generally I recommend only training the mistakes.
After I adjusted the spam strategy slider back to mid-point and re-set the statistics, accuracy jumped to 100%. This morning, after two (or three, I don’t remember) false positives, accuracy dropped to 99.3%.
I’m sure I’d be better off emotionally if I just stepped back and stopped watching the numbers change…but sometimes it’s hard to just leave things alone.
Thanks. Please make sure that you send diagnostic reports to the reporting address rather than to the public forum. (I’ve deleted it from the forum for privacy.)
I do not see any messages trained as good since you updated to the new version.
When you were using the old version this morning, there was only one false positive (HelloFresh). The Capital One message that you trained as good had not been processed by SpamSieve at all. Perhaps it was moved to Junk by a server junk filter before it got to your Mac.