@Undo I would get a diamond somewhere... But the thing is that I'm not active enough anywhere to get one, and while I do do quite a bit of moderation on many sites, sadly its rep that the voters look for
Also, I'm writing up a quick Google doc with my thoughts, is it fine if I get it to you sometime in the next hour for you to send (or not to send ;) to someone
[ SmokeDetector | MS ] Chinese character in title, mostly non-Latin body, mostly non-Latin title, pattern-matching website in body: 携帯電話ジャマーを入手する方法はなんですか by camera on askubuntu.com (@ThomasWard)
[ SmokeDetector | MS ] URL in title, blacklisted website in body, blacklisted website in title, pattern-matching website in body, pattern-matching website in title: getmusclebuildingbase.com/astoria-vr/ by ebbMary on apple.stackexchange.com
Quick...in less than 50 words...someone describe keybase.io to me and why I should a.) trust them and b.) trust they aren't going to disappear in the next X months
if the functionality is to be integrated to SO proper (is that what the document is for?) then I think you need to think about what that system will look like, rather than what the current system looks like
I don't think SO is going to take code from Github and run it on their servers without a process on their side
so the first question is then how do you classify a rule as confident, and make sure it only gets modified by trusted users, with adequate testing before releasing an update to production
off the top of my head, what I would suggest is to keep the current system as a test bed for rules, and only merge rules for which you can display test results over some amount of time
merge to the SO copy I mean
then, "being trusted" is not enough, you need to do some paperwork to show that the merge is according to process (i.e. it ran for x hours and produced y hits and z false positives)
Also, just make a big deal out of any false positive we accidentally flag. That's served us well in the past, no reason it can't continue working. For bonus points, automatically retract flags if the post isn't showing a downward-trending score after x minutes.
@tripleee As a ground rule, we'd want to limit the system so that it can only cast x % of the flags required to actually nuke something. That mostly eliminates the alcohol factor.
@BrockAdams Sure. I'm talking about "we cast a flag ten seconds after it was created, and five minutes later no other humans have flagged. Something is weird, we're bailing"
still, to me, the most obvious risk here is that there is only the live code base -- if you were to start rearchitecting some part of it, what's to say it won't suddenly start doing weird, uncontrolled things
so what I guess I'm proposing is to run a test and a prod version in some way or form
probably what I'd like to see is a single live code base but with checks to disable the auto-flagging and sane and reasonably strict rules for when it can be re-enabled
so in the terms of the linked document, "vulnerability: code base could be destabilized by mistake or malice" (not really a "vulnerability" but a "risk") and "current mitigations: somebody with privileges could notice and revert back to a stable commit, hopefully sooner rather than later" and finally "possible improvements: define a process for when the code base is declared stable enough to enable auto-flag" maybe?
implicitly then, flagging would be disabled for a while after a commit goes live, I suppose?
or you'd have two instances, one with vetted, stable rules, and another a test bed with flimsy heuristics and new rules still in development <-- this would be attractive for other reasons, too
[ SmokeDetector | MS ] Bad keyword in title, bad keyword with email in body, blacklisted website in body, repeated URL at end of long post: ETL TESTING ONLINE TRAINING by Umayaan4411 on stackoverflow.com (@dorukayhan)
tpu- by Glorfindel
[ SmokeDetector | MS ] URL in title, blacklisted website in body, blacklisted website in title, pattern-matching website in body, pattern-matching website in title: www.healthyorder.org/apexatropin/ by Gret19 on apple.stackexchange.com
tpu- by Glorfindel
[ SmokeDetector | MS ] URL-only title, link at end of body, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: goldenhealthcenters.com/astoria-vr/ by user630754 on askubuntu.com (@ThomasWard)
indeed. while not inherently bad, the user in question who was posting it was saying "You could also try [link]" as a solution to a question on how to cut/splice audio under Ubuntu. And it's just an ad company. (Such links should be purgified)
