I'm trying to make sense of Pigeonhole

[IT_Architect]

I know it can't be complicated unless people intentionally make it that way:
1. Is this anything more than what Microsoft Outlook has had for 20 years where you can make up rules and based on the E-Mail's headers, source, destination, subject, etc. they can be sent to the folder of choice or perform actions?
2. Is the Interface to set up the rules simple like Outlook or it is something only a geek could love?
3. It appears it integrates teaching SpamAssassin like we set up now manually?
*So basically, it makes RoundCube more like Outlook Web Access?

Thanks!

 

[mxroute]

For basic usage it's just as easy as any more common filtering system. It can accept regex, and that can open it up to much more complex usage. The most common user issue is misunderstanding. People think with context, filters don't. They think they set it to do one thing, but they actually set it to do something else because they don't understand how literal it is.

It's 100% always misunderstanding that leads to issues. Not bugs, though your users will try hard to tempt you to consider their misunderstandings to be bugs, especially if you can't decipher the logic on their behalf. I've literally refunded and sent customers packing for harassing me over tickets, repeatedly demanding that I consider their misunderstandings as bugs, even though I have no intention of contributing to the dovecot project's code (not my specialty).

 

[IT_Architect]

Thanks!
It sounds like something I don't want. I installed it because the follow-on instructions show it to be capable of teaching SpamAssassin. I didn't notice any changes inside of RoundCube. I was hoping it would add functionality as seen in Outlook since 2001. I did not see that. One certainty and that is users don't work with regex. The follow-on instructions I was hoping could enable teaching SpamAssassin aren't even close to being valid. It references paths that do not exist, and after you fix them, it requires includes for a compile it doesn't find, after which you find configs the reference files with slightly different names, none of which are anywhere on disk. So that part is unusable and standard SA learn technique is what I will need to use, and probably a better way anyway.

 

[mxroute]

Thanks!
It sounds like something I don't want. I installed it because the follow-on instructions show it to be capable of teaching SpamAssassin. I didn't notice any changes inside of RoundCube. I was hoping it would add functionality as seen in Outlook since 2001. I did not see that. One certainty and that is users don't work with regex. The follow-on instructions I was hoping could enable teaching SpamAssassin aren't even close to being valid. It references paths that do not exist, and after you fix them, it requires includes for a compile it doesn't find, after which you find configs the reference files with slightly different names, none of which are anywhere on disk. So that part is unusable and standard SA learn technique is what I will need to use, and probably a better way anyway.

To that end, I went down that path back in 2014 with MXroute. I set it all up for customers and we ran sa-learn automatically based on user reports for something around a year if memory serves. The truth is, the learning that it uses can't keep up with today's trends. At best, it resulted in false positives. At worst, it did nothing. But in both cases, the CPU overhead added up over time to create a lot of resource usage and the only gain was that people "felt" like they were influencing their filters. If you plan on continuing down that path, and if I may offer any advice, I'd focus it toward rspamd's neural learning instead. I don't know that I've seen better results with it, but resource usage is a fraction of a fraction. Though I'm just doing it at root level myself, haven't implemented any user-level training.

Now rspamd's public fuzzy server, that's some good stuff that's constantly being trained because it's always being fed new spam. Honestly, you might find just using that to be equivalent to the best results a user might get from training it themselves.

 

[IT_Architect]

I've had fabulous results when I train someONE in the organization how to train SpamAssassin. I have not tried rspamd. I have read about the pros and cons of each in the DirectAdmin help, but that is about all. Since I have had such great results with the training, I went with it. Thanks for your thoughts on rspamd.

I got Pigeonhole working with DirectAdmin. It actually was, but since it is not documented, I learned about how it worked at https://help.poralix.com/articles/creating-autoresponder-via-webmail-in. In a way I'm glad that their instructions of how to set it up Pigeonhole to train SpamAssissin are bogus because at the virtual user level might not be the best for training. I've been having one person do it for the organization that is trained not to train it on things that could go either way.