I think it'd make a great SoC project -- agreed! --j.
Warren Togami writes: > (Below is just a strawman idea for a project that I think would be very > beneficial to the future of the spamassassin project. Please add your > ideas or possible implementation details in reply. Then later we can > summarize it and put it on the Wiki.) > > Nightly mass check is how we check rules in the SandBox and adjust to > the ever changing behavior of spammers. But there are many problems in > the current process which limits its current effectiveness. > > Problems > ======== > 1) It is only really usable on Linux/Unix hosts, or maybe Cygwin with a > bunch of effort, although not easily with any Windows clients. Even if > you are a Linux user, it is *TOO HARD* to setup and get it to work properly. > 2) As a result, most if not all people sorting corpora and participating > in nightly mass checks and corpus scoring are somehow related to a > single demographic: computer hackers. > 3) We don't have participants who are representative of normal users. > These are people that use computers for reasons other than their work or > hobby. > 4) We don't have various languages represented in the corpus, especially > Asian languages. > 5) Normal users are less likely to understand the sorting rules and > require training. Maybe their results would be less trusted. > > Proposal > ======== > We need a way to make nightly mass check easily accessible to normal > users. They need easy to use software to do mass checks and submit > results. They must be properly trained on the sorting rules. Our > project then needs some way of tracking the level of trust of these > growing number of submitters. > > I envision that generally hackers that care about spamassassin will urge > their non-hacker friends to use this software as part of their daily > e-mail. It is easy to convince people about the social benefit and how > you can volunteer some time to help the rest of the world. > > I think it would be sufficient to have a few dozen participants from > different demographics, regions and languages in order to improve > spamassassin. After this more accessible mass check software and > supporting project infrastructure is ready, we could do a call for > volunteers where our community can go out and find people in these > varying demographics to participate. Our existing community of hackers > and sysadmins can train individuals in corpus sorting and get them started. > > You may think "this is crazy, why Windows?" > The reason is this system *MUST* be easily accessible to normal users. > > Requirements > ============ > 1) MUST be able to run on Windows, where most normal users are. > 2) MUST be able to read local Outlook, Outlook Express, and Thunderbird > mail folders for ham and spam. > 3) MUST be able to submit results to the spamassassin project. > 4) MUST be very easy to setup and use, point and click. No editing of > text files. > 5) MUST document an easy to understand guide for normal users to learn > the corpus sorting rules. > 6) MUST devise some kind of improved accounting system that allows > different levels of trust in submitted nightly mass check logs. > > Possible Implementation Details > =============================== > 1) Implementing this really wouldn't be all that hard because you would > use existing components like perl, spamassassin, ssh, and rsync. > 2) You would need to tie them all together using some toolkit to make a > frontend like gtk+ or qt that works in Cygwin. From the frontend you > choose mail client folders to read for HAM and SPAM, and schedule the > nightly mass check time. > 3) You might need some kind of service or applet visible from the > systray in order to do the scheduled nightly mass check. > 4) Use an "InstallShield" type click-thru installer, which seems to be > the standard on Windows. They shouldn't need to make *ANY* choices to > download and configure Cygwin. It should just dump everything needed > into a single folder that contains this mass check bundle. > 5) The use of existing FOSS components and a cross-platform toolkit like > gtk+ or qt would allow this to build on Linux too. > > Thoughts? > > Warren Togami > [EMAIL PROTECTED]
