@hichris123 oh :P

pulled

[ SmokeDetector ] Link at end of answer: How can I switch an application to a different playback device on Windows 10 by user2597747 on superuser.com

@SmokeDetector seems legit

@SmokeDetector ignore-

@Undo You should post that to TL; no doubt a number of mods would have things to say about it.

And CM's, for that matter.

Eh, good idea.

Maybe title it RFC, following the storied tradition.

k

I have zero experience writing these, feel free to do whatever you want to it

@Undo you should probably also change it so that people can't just edit... use suggested edits instead

@ArtOfCode Not seeing that

> can comment

@Undo try comment

there

That's it. I'm suggesting now, apparently.

huh. I have no idea how Google Docs works.

Yes, this comes with commenting.

@Undo Web Applications :P

this is... certainly interesting

I am skeptical re: CM's reaction to this, esp. after the rough week they had. They might see it as a form of sockpuppetry, one user assuming the identity of another (with their consent, etc). But that would be a concern with any form of proposal.

@Doorknob As I see it, this would essentially be a way to send signal to SE that "Smokey things this is bad, and we're confident enough in that to knock the number of 'organic' spam flags required down by one or two"

@Normal Possibly, but if we do our homework we have a better shot

Concerning the implementation, an alternative is to implement !!/flag-spam command or such, available to very few users.

Which would turn loose other people's spam flags?

Which uses someone's token but is issued manually.

@Normal Valid concern, but they have seen metasmoke data before and have even been known to comment on how accurate it is.

My dream for this would be for SE just to implement it into SpamRam. But the likelihood of that is... quite slim.

@ArtOfCode That's your TL-private affairs. :)

@Normal That gets too close to puppetry IMO. Having a machine behind this, operating on a very specific set of artificial restraints, is safer than having a person triggering it.

Basically, I'd like to keep people as far away from this as possible. Less chance for abuse.

Good point.

It might change how people use tp's (probably make them more common), but that's okay

Where 'abuse' means 'actual abuse or alleged abuse'

Thought: let's start setting out more detailed requirements of a post before flagging it.

>= 1 tp.

Post score?

Poster rep?

Post score <=0, owner rep =1, obviously.

Is there some way we could integrate past data into determining how likely a report is to be accurate?

@ArtOfCode That'd probably be manually determined

are tps already privileged?

@Normal Which means from bodyfetcher. Which would make sense -- if the post is still around, the bodyfetcher could get to it.

@Doorknob yes

i.e. if a site has seen lots of tp's X reports before, another X report is perhaps more likely to be true

@hichris123 they are?

if so, perhaps only a "super-privileged" tp should trigger this. I wouldn't trust... uh, certain users from other chatrooms to flag on my behalf

I'm almost certain I've never been added to the privilege list...

Yeah, but we've been somewhat lenient on giving people those privs

@ArtOfCode You're a mod. :)

@Doorknob Would you trust it if that was used in conjunction with a bunch of other filters?

@hichris123 Oh, that's included? Neat.

(which you would set up)

(mods are auto-privileged)

How do those last two lines look?

(do we need more parentheticals?)

@Undo like post score and user rep? I'm thinking specifically of that one incident where someone from a SO chatroom kept blacklisting a user who was only borderline spamming, definitely not the "baba" etc. stuff that's absolutely 100% spam

@ArtOfCode Yeah, I wrote that in after realizing it takes too much effort to add all interested moderators to the list. :P

Hmm, what's going on in here?

heh

@Undo I think <0, as you had it there, is better. At least one human flag must be there. Or maybe even two.

@michaelpri We're thinking of doing something new and dangerous.

@michaelpri We're specifying an auto-flagging system :)

that user happened to have 1 rep and obviously nobody upvoted the low quality answers, but I wouldn't flag the answers as spam if it's not the "obvious" kind

@Normal True, and by the time it gets tp'd it should be <0.

@Undo Sounds fun :P

See the pinned thing in the sidebar.

Ah, very interesting...

@Doorknob Restrict it to questions at first, maybe?

And have manually reported stuff out of bounds, too much potential for abuse.

definitely

Hmm, a quick solution to false positives would be to restrict it to drupal/AU questions.

I don't think those have many fps.

At first, I'd like to have a whitelist of reason/site combinations that users can even choose.

@Undo Have a list of reasons that we're confident in; then let users pick a subset of those to enable flagging for.

Maybe have an automated check in metasmoke that makes sure whatever filter a user chooses has >x% tp rate (or <x% fp rate).

@ArtOfCode Right

questions only, user rep 1, score ≤ 0, select reasons, requires {n} tps by super-trusted users?

@Doorknob Sounds like a pretty tight requirement.

Which is good.

that's intentional

On a quick search, I don't see any fps on Drupal in the past 10 days. metasmoke.erwaysoftware.com/…

yeah

Reasons are kind of fuzzy. "Bad keyword" is the most common cause of report, but sometimes it's "garcinia" (100% spam), sometimes "brain power".

We can always open it up later; the goal is to get this past CMs.

@Undo can you run that on the metasmoke DB? Select posts with user rep 1, score <= 0

I don't think I have either of those.

@Normal it could be separated into "super-spammy keywords" and "signaling keywords" that have fps

@ArtOfCode Rep and score aren't reported there. But all automatic true positives fit these.

Ah right

@Normal We could send why data to metasmoke and filter on that.

Good idea, regardless of this project actually.

yep

Especially if you add a search field, search by "why".

hey Google Docs, I don't need you to email me reporting that Undo accepted my suggestions when I still have the tab open :P

lol

blacklisted website in body should definitely be one of the reasons

@hichris123 The likelihood of us maintaining the level of access we have now is even slimmer.

Hmm?

@hichris123 Regarding how fast we can change regexes and adapt.

Smokey's advantage is that we can change things within minutes without really breaking anything. As soon as it becomes a system thing, there's something at stake and the rapid iteration becomes much harder.

Oh. That is true.

@Undo Does metasmoke have data on how accurate different reasons are for different sites?

@michaelpri It's there, yeah

Kinda have to extract it, but all the pieces are there

@Undo We should probably also review some of the feedback on metasmoke - I noticed some of the fps aren't really fps (must've been a mistake) and some of the more borderline posts were tp-ed.

How would we do that?

[ SmokeDetector ] Repeating characters in answer: Uniqueness of coefficients for shape functions of higher-order finite elements by Radek Martinez on scicomp.stackexchange.com

[ SmokeDetector ] Few unique characters in answer: Can F-cycle substitue FMG for update of existent solution? by Radek Martinez on scicomp.stackexchange.com

@Normal OK, here, have some negative reputation points.

And some delete votes from me on Math.

@ArtOfCode antirep

@ArtOfCode On Math, it may well work for him...

The automatic flagging would be only for spam, right?

I suppose so. The RFC says so from the beginning.

Ok, because I feel like other types of flagging (besides offensive) would be incredibly unreliable.

@michaelpri Right

offensive == spam, in my book

I made a few changes to the specs. The only major one was to include this as well

Offensive is much more unreliable, though. Someone posts on SU or AU in the state of extreme frustration, and has a swear word in it, and gets justifiably downvoted (for this or other reasons)... still not enough evidence for a flag.

I'd limit it strictly to spam, as in spam.

Me too. Several of the one off swear words can be easily edited to remove such an infraction.

@Andy Maybe, but that's architecturally hard.

@Normal There are some words, that should clearly point to being flagged as offensive, but we could set up filters to only autoflag on certain words or patterns of words.

The best we could do is a cronjob running every 5 minutes, but then a post could still be flagged anywhere from 5 to 10 minutes afterwards.

Which is quite a long time, really

So, a valid question we need to answer then: What is an acceptable time frame for spam to survive?

And, how long does an average spam post survive now?

Preferably, <5 minutes

In a perfect world, we want it gone instantly.

@ArtOfCode Ideally 1-2 minutes, which I believe we're capable of

@Andy Greatly varies. The bitcoin spam (questions!) lived painfully long this past night.

@Andy I've seen them survive >5-10 minutes during that magic 4-12 UTC timeframe where no one is around

I'd suggest no automatic flags for 5 minutes of the post's existence.

Give users a chance to handle it, then autoflag if necessary.

I agree

TIL: Google Docs kicks you out of the doc if you idle to long

That's kind of annoying

5-10 minutes isn't that long.

especially the lower end of that

really? doesn't do that for me

^

I just kicked me out. shrugs I blame Google.

[ SmokeDetector ] Repeating characters in title, title has only one unique char: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa‌​aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa by Matjaž Jerman on stackoverflow.com

Is this RFC what we are sending to SE? Or is this just to get initial ideas of what we want to present to SE?

@Andy Either, we'll see what we think of it in a few days

"Present" probably just means "Hey Shog, we want to do this thing, is that okay? We've done our homework, <link>"

Ok, if this is what we are sending to SE, I think the introduction needs more details on what both Smokey and metasmoke are and do. We know what they do, but we want the person receiving it to be clear on what they do too

I'll put some more work into it later. Right now, it's time to make something to eat. I'm to much of a wuss to go outside when it's -5F with windchill out, so I have to get creative.

Overall, it looks better than many RFCs I'm handed at work. If I got this in the office, I'd buy you lunch for actually doing your homework

I like this idea:

If nothing else, it might bring more structure to the big messy list.

The trick is that renaming a reporting reason would mess up Metasmoke stats, which is undesirable.

However, I recall that Metasmoke strips out parenthetical comments to reasons. Maybe we could provide these. "Bad keyword in body... (strong)"

Hm.

@Normal Yeah, but we do that at intake, so there would be no way to distinguish between them later

We could put that in why data, though

Or just a separate numeric field for Metasmoke, the strength of signal.

that'd work

Restart: API quota is 8153.

@Undo Metasmoke = 503

This might hurt the credibility of the project a little...

heh

I might just move it to a dedicated server.

It's back up

I really shouldn't be this cheap with these kinds of things. It's not like I have any other expenses.

Apparently, building on Rails isn't cheap, hardware-wise.

It's actually fairly cheap, the issue is that my hardware is far cheaper

My Apache conf files are huge and numerous.

Now that you have RFC out, a Kickstarter campaign for hardware comes next.

lol

"make it more dynamical"

That's what your server needs to be, apparently.

D'oh, that's it! I need Ubuntu 14.04 Dynamical Server.

14.04 Dynamical Tahr

> A routine of feeding during the morning followed by a long rest period, then feeding in the evening, constitutes the tahrs daily routine. Tahrs are not generally active or feed at night

What kind of server is that, feeding and then resting the rest of the day.

> A routine of rebooting during the morning followed by a long working period, then rebooting in the evening, constitutes metasmoke's daily routine. Metasmoke is not generally active or rebooting at night.

@Normal A dynamical one.

Shiny buttons.

MOD ABUSE!