Charcoal HQ

M.A.R.

6:00 PM

@terdon the world is a diverse place

DJMcMayhem

As much as possible, or up to a certain point?

Magisch

If a spam post gets like 4 views and then dumpstered, thats nice

doppelgreener

@Undo Remove spam.

Magisch

Having 1 or 2 charcoal members look at it for feeding back is fine, we signed up for this and we have the tools and the preparation to not be caught off guard

M.A.R.

@doppelgreener this

People not seeing it is the side effect

Magisch

6:00 PM

Ideally, no unsuspecting site member has to see the crap we delete

Jake Symons

@Magisch You make it sound like we've had military training in spam combat :)

doppelgreener

@M.A.R. right.

Undo

@doppelgreener Must be deeper than that, though - if that were our only goal, autoflagging wouldn't exist at all. Everything gets removed... eventually.

M.A.R.

@JakeSymons you didn't? Noob

SmokeDetector

[ userscripts ] continuous-integration/travis-ci/push success on 3d8c9bd: The Travis CI build passed

M.A.R.

6:02 PM

@Undo autoflagging happened because we wanted it removed as fast as possible

John Dvorak

just say "as soon as possible" like a normal person would...

2

M.A.R.

@DJMcMayhem what acronym?

So, anyway, goal is "removing spam as fast as possible", put more elaborately

Whether people are happy we do it for them has never been an issue up to now

We took it as granted. Spam tastes awful

doppelgreener

Support the network's health and integrity by removing spam so that the network can go about its business unblighted.

To that end:
- We have a trained bot to that extent because it helps identify spam.
- We have autoflags because our bot can help make spam disappear more quickly when it has reason to be confident it has found that spam.
- We have Smokey subscriptions in other chats so communities can have oversight & get involved in removing spam spam.
- We want to remove spam faster because the more spam there is and the longer it stays, the more health is compromised.

Shog9

@terdon opt-in is a terrible solution, and I don't think anyone should seriously consider it at all. Most of the larger spam groups don't care which site they're on - the account system and rate-limits make it extremely convenient to just go site-to-site, dropping spam wherever it is convenient. As a result, "opt-in" as a compromise is just a dishonest way of punting the decision until the next wave of CJK / movies spam at which point it becomes an opt-out system.

John Dvorak

@M.A.R. you can achieve that by six-flagging anything that might be spam. Bad idea.

Shog9

6:06 PM

But... The reasons behind wanting opt-in are varied and worth discussing.

terdon

@Shog9 I don't feel particularly strongly about opt-in, personally, I just get the arguments behind it. But we're still talking about opting in to 5 autoflags, not opting in to smokey in general.

M.A.R.

@JohnDvorak fine. "Removing spam as fast as possible, acting on posts not alright but not spam either, and not touching alright posts"

quartata

@Shog9 this has been happening recently, where you get a spammer that just makes 2-3 network profiles and spams every single site until they're gone

Magisch

I think having spam gone asap with as few user eyes on it as possible is best

quartata

not as common but definitely a nuisance

Magisch

6:08 PM

because it minimizes network disruption due to spam

doppelgreener

I'm choosing that goal phrasing because 5 flags fights spam but ignores the "support the network's health and integrity" part of the goal because it has potential to compromise its health and integrity.

Shog9

@quartata not just recently

FWIW, there are at least 4 separate spam protection systems built into the Stack Exchange software itself, none of which are thoroughly documented publicly and none of which have anything approaching real transparency. Those are the real alternatives here, not that any of you have enough information to do a fair comparison of them with SD.

Magisch

after all the reason we get rid of spam is because it's unsightly, off topic and disruptive

quartata

well most egregiously recently. They always have the same username which is why I notice

M.A.R.

@Magisch but . . . Two eyes and two hands is too few

quartata

6:09 PM

I don't know what accounts share IPs

terdon

@Shog9 Well, we do know that SD at the very least improved them enormously. Just by comparing the previous status quo to what we have today.

Magisch

@M.A.R. they're supported by the cumulative experience and statistics we've gathered on 85k spam posts

terdon

I mean, "I remember back when spam was a problem" say the old timers.

Magisch

spam is still a problem, we're just better problem solvers now

M.A.R.

For hands and two brains are much less likely to cause trouble, especially from people from around the world.

doppelgreener

6:10 PM

I am actually in the camp that doesn't care whether spam gets deleted in 5 seconds or 10 seconds or 1 minute or 5 minutes. Anything under 10 minutes is awesome to me.

Magisch

it's an ongoing problem, and constant adjustments and reactions are necessary

M.A.R.

@Magisch right, I was getting to that

terdon

@Magisch I mean it's a problem that stops at charcoal's doorstep when it used to be a serious problem for the sites downstream of you.

Shog9

@terdon my point is, there's a lot of hand-wringing about SD's effect on individual sites, but it's mostly blind to the fact that when I get pinged on a Friday evening about a spam wave, I'm not worried about false positives or user transparency or being nice to borderline promoters... My concern is finding the biggest hammer as fast as possible so I can get back to drinking and letting the collateral damage sort itself out on Monday.

5

Magisch

@terdon which I count as a positive since we're the masochistic people who signed up to keep this shit away from where people really don't want it

terdon

6:11 PM

@Magisch Of course it's a positive! What else could it be? You really need to get this idea that people are against SD out of your head.

David

@Undo encase it got lost in the transcript. the white list blender domains chat.stackexchange.com/transcript/message/43243851#43243851

doppelgreener

Here's what I understand about the spam business model:
- Spam arrives.
- The spammer sends a link to their boss.
- The boss checks it later and if it's still around, good job, the spammer gets paid.
In that business model do we actually care whether spam gets deleted in 5 seconds or 5 minutes? Either way by the time the boss checks it, it's gone and the spammer doesn't get paid.

The problems are when spam sits around for half an hour or an hour.

Shog9

@doppelgreener near as I can tell, that's not entirely accurate

terdon

@Shog9 Sure. I've had to do similar things on AU only with a much smaller hammer.

Magisch

@doppelgreener thats not how it works

the checking happens very quickly after

David

6:12 PM

@Shog9 is that how you CMs work? O.o

2

Undo

@doppelgreener Almost certainly an automated check

terdon

SD is one of the best things to come out of the network in ages.

Shog9

That is, a lot of spam isn't actually concerned about the link itself. They want keywords on Google. So they verify that the link shows up in Google.

They're after our SEO, not our users.

doppelgreener

Ok, I see.

quartata

@terdon (see SFF)

Magisch

6:13 PM

to that end quick deletion helps

Undo

@Shog9 They'd figure out the nofollow pretty fast, wouldn't they?

Shog9

@Magisch yes

Magisch

if we get it before its keywords get indexed, we're golden

and the spammer has just wasted his time, and will find that out

quartata

@Undo the links still show up in SE custom search and what not

Shog9

@Undo noindex is what is needed

quartata

6:13 PM

yeah

Magisch

Now they might move on to another site with less protection, and file stack exchange under "not worth the bother"

quartata

@Magisch yeah that's not happening

terdon

@quartata Oh, I don't take that user seriously.

CalvT븃

So really our business model is... beating Google

quartata

@terdon *users

I do though, I feel obligated to take them all seriously

terdon

6:14 PM

But, actually, do see SFF. The consensus is pretty clearly pro SD

Magisch

@quartata It has happened heavily. Do you have any idea how hard other sites are drowning in spam?

Mad Scientist

@Shog9 couldn't you just noindex all new posts for the first 15 minutes only?

quartata

@Magisch except SO is #70 on alexa

Magisch

what we're dealing with here is the few holdouts after spamram and SD

quartata

it's getting spam

Undo

6:14 PM

@MadScientist Internal database structure makes that difficult

Shog9

@MadScientist heck, we wouldn't even need that - the SE system actively feeds new posts into Google.

Magisch

minecraft forums is not in the alexa 100 and we didn't have spamram there

Shog9

Of course, that's usually an advantage. So kinda cutting off our nose to spite our face.

quartata

@shog To be honest, this is the first I've heard of "Smokey's" thresholds and operating parameters. You may think this is good, in that good automation should be seamless and happen auto-magically, however, this is a site moderated by the community. The moment you take that interaction and engagement (no matter how sullied) away from users, you're taking away from the concept of Stack. Spam is not such a problem here on this site that we want this feature. — Möoz 15 hours ago

Magisch

every weekend moderators would spend hours deleting spam and accounts in seconds to just have them return harder

Shog9

6:15 PM

Yeah, please do ignore SFF

as I said above, there's absolutely nothing interesting about the spam there.

terdon

@quartata Yes, that's um. A special user. It's also just a random comment on a meta discussion, not representative of the site and not something you should (or even can) use to gauge reaction.

Shog9

y'all want to make Mad & terdon & especially JG happy. They represent a lot more users and a lot more spam.

terdon

Yeah, last time I was here there was head rubbing!

quartata

@Shog9 well to that end I think first things first is the RSS right @Undo?

we've got the post search one currently

figure we can steal the view

Undo

@Shog9 Not sure how to make JG happy without an opt-in process. Maybe do opt-in for SU?

M.A.R.

6:17 PM

I was composing a philosophical and long message, but the conversation shifted away already. Dang it.

Undo

@M.A.R. Brings back memories of Ye Olde Meta Towne Hall

thesecretmaster

@quartata We want rss of autoflagged tps?

quartata

I think that was it, yeah

...which come to think of it

I think they can already get that

Shog9

@Undo as I said above, read past the proposed solution and look at what's making him concerned.

Magisch

We should start by addressing JG's 4 concerns outlined underneath

to wit:

> A. Does optimising for shorter lifetime for spam reduce the chances of them getting picked up on search engines?

B. What are the benefits for community outside the outcome of spam getting deleted faster?

C. How does this compare in time to deletion to fewer auto flags, or more precisely, what's the minimum number of flags at which auto flagging is useful?

D. As a downside, does this discourage people outside Smokey from flagging spam (since they don't get into the habit of it)?

M.A.R.

6:19 PM

@Undo aye

Shog9

To my evaluation, that is three things:
1. Reliability of flagging

2. Moderator awareness of flagging

3. General efficacy of the system

Undo

> A. Does optimising for shorter lifetime for spam reduce the chances of them getting picked up on search engines?

Shog9

Address those and opt-in will not be necessary.

As a bonus, addressing those helps every other site

Whereas opt-in doesn't.

Undo

@Shog9 Ah, gotcha. Reliability we have numbers on. Moderator awareness is well addressed with technical changes, like yours and Mad's. Efficacy I'm less clear on.

doppelgreener

@Undo On (A), I would make a point of focusing on: do the 3 vs 5 flag times (i.stack.imgur.com/Kph8d.png) reduce those chances? I see a graph but I can't actually tell what the numbers are in it -- I'm looking at a tiny section of the Y axis. What are those numbers and how does the difference between 3 & 5 affect spam?

At some point, we'll be reaching a "fast" that is "fast enough". (I don't know what it is.) Any speed beyond that is not actually necessary.

quartata

6:21 PM

ok so just to clarify @ArtOfCode the only parameter /posts/search takes is the site

Undo

Oh!

quartata

we need to expand that

Shog9

@Undo finding a way to demonstrate that faster deletion == less spam would be a big win...

Magisch

@Undo efficacy as in clearly explaining why it's good to have 5

Undo

Maybe this is it. The core of everything: If spammers are going for our SEO as Shog says, they might be banking on those TTD outliers. Five flags makes those go away, apparently.

2

SmokeDetector

6:21 PM

[ SmokeDetector | MS ] Username similar to website in answer: In Ansible, How to connect to windows host? by Devopstechie on stackoverflow.com

David

alright finally posted my answer to the meta Q:

Magisch

@Undo this

David

0

A: A machine can flag spam automatically. Can it do better?

Note of warning I have not read all of the answers here. I'm sure what I wrote is duplicated somewhere. I like smokey, it works amazingly well. Yes I would trust it to fully automatically delete spam. I'm going to point out some "issues" and then speek to them (we are on meta after all). The...

Undo

Posting a dozen things in hopes of getting one that lasts two minutes is worth it to them.

Magisch

Spammers gain nothing from spam thats gone after 10 seconds

Now if it stays for 20 minutes, even if only 1 out of 100 posts they make does that

quartata

6:22 PM

@Undo speaking of which

Magisch

it might be worth it for them

Undo

Make them post a hundred or a thousand things to get that one outlier, and maybe it becomes less lucrative.

SmokeDetector

fp- by doppelgreener

Undo

Now, that's all speculation. No data.

Magisch

@Undo It's fact based speculation though

We know spammers care about our SEO

so the reason they spam is for SEO

If we can deny them that then their incentive lessens

doppelgreener

6:23 PM

@Undo This is good.

Undo

well, here's the way you get to it: (1) Spammers are here for profit. (2) Do posts deleted within ten-twenty seconds get them profit?

If (2) is no, then they must be after the outliers.

Kill the outliers and you kill the business.

Magisch

kill the business and people will bother less to spam

Shog9

I gotta get to some other stuff today, but I want to leave y'all with a bit of encouragement: I have these exact same arguments within the company. Trying to demonstrate that spam is harmful on more than a gut level isn't easy or fun - but it's worthwhile in the end, because you get people willing to work with you instead of against you. Thanks for all your hard work, and good luck!

Undo

@Magisch And Shog gets to drink uninterrupted.

Thanks for taking time, Shog. Means a lot to us :)

2

doppelgreener

Right now the message is: "We want to go fast. We currently go really really fast! We want to go faster. There might be some injuries along the way but we're pretty sure there won't be and we want to go faster and going faster will be worth it." The community is looking at that and going: "... But why would you do that? Fast is already enough."

5

terdon

6:25 PM

This.

Undo

Okay, honest question: Where do I go to get someone to pay me to spam? I want to see how this works internally.

thesecretmaster

@quartata It should take site, feedback type, from_date, and to_date. I'm going to make a PR to add autoflagging now, and add a rss feed too.

John Dvorak

@doppelgreener PR is hard, m'kay? :P

quartata

@thesecretmaster thanks that was fast

doppelgreener

@JohnDvorak no doubt on that

but that means this is the direction we should go in:

5 mins ago, by Shog9

@Undo finding a way to demonstrate that faster deletion == less spam would be a big win...

4 mins ago, by Undo

Maybe this is it. The core of everything: If spammers are going for our SEO as Shog says, they might be banking on those TTD outliers. Five flags makes those go away, apparently.

Mad Scientist

6:26 PM

@Undo so we should encourage more spam to save Shog's liver?

doppelgreener

it's not "go faster, because... well, faster". It's "go faster because it means we'll hardly even get spam in the first place because everyone will just give up" or something.

tripleee

@Undo Brian Krebs wrote a couple of books about that perspective, mainly email spam though

thesecretmaster

@doppelgreener This is a really good argument. I was actually kinda against going the full 5 flags, but this finally makes sense to me.

doppelgreener

@thesecretmaster I am in exactly the same boat as you there.

John Dvorak

+1

terdon

6:29 PM

Yes. Until now the argument has largely been (or come across as) speed for speed's sake. This makes much more sense.

doppelgreener

@Magisch that also answers B & C here

11 mins ago, by Magisch

> A. Does optimising for shorter lifetime for spam reduce the chances of them getting picked up on search engines?

B. What are the benefits for community outside the outcome of spam getting deleted faster?

C. How does this compare in time to deletion to fewer auto flags, or more precisely, what's the minimum number of flags at which auto flagging is useful?

D. As a downside, does this discourage people outside Smokey from flagging spam (since they don't get into the habit of it)?

B) less spam arrives in the first place. C) see B.

SmokeDetector

New metasmoke user 'yivi' created

John Dvorak

@terdon I always felt it as speed for fewer spam's sake, but I agree it's good to have it explicit

SmokeDetector

[ SmokeDetector | MS ] Bad NS for domain in body: Picture can't be linked due to Wordpress Plugin by Rita Schweigli on stackoverflow.com

quartata

@Undo t = -2.6374006236321903, v = 68.88519325013864, tcdf(t, v) = .005159

(assuming I did it right lol, one second)

just whipped up a quicky Welch t-test in Python because I couldn't be bothered to open up Octave

that's between 5 flags and 4 by the way

SmokeDetector

6:33 PM

fp- by Glorfindel

John Dvorak

How do I read these numbers?

terdon

@JohnDvorak The difference is between simply removing it faster and causing less to be posted in the first place.

Andy

Anyone with 10K on Chemistry? If so, can you tell me the reason this was deleted?

quartata

@JohnDvorak t is the test statistic, v is the degrees of freedom. The last one is the p value (i.e. the one you care about. small is good)

SmokeDetector

New metasmoke user 'John Conde' created

John Dvorak

6:35 PM

@quartata so... you have formally proven that adding a flag almost definitely reduces the spam lifetime? PogChamp

quartata

yes

Glorfindel

@Andy @Loong might be able to help.

quartata

but people question that you see

it's important

Loong

@Glorfindel checking

John Dvorak

Oh, okay :-D

tripleee

6:36 PM

in SO Close Vote Reviewers on Stack Overflow Chat, 4 mins ago, by SmokeDetector

[ SmokeDetector | MS ] Bad NS for domain in body: Picture can't be linked due to Wordpress Plugin by Rita Schweigli on stackoverflow.com

weren't these supposed to not go to SOCVR?

quartata

no

bad ns is no longer experimental

tripleee

ohhh

John Dvorak

How expensive would it be to have reasons clickable in reports?

SmokeDetector

New metasmoke user 'mooseman' created

quartata

On the other hand, 3 vs 4 is much weaker (p = 0.03)

That's still statistically significant, but much much less so

So really this tells us what we already know: we want 5 flags

John Dvorak

6:39 PM

I'm surprised how close 2 vs 3 is

Loong

@Andy Manually deleted by moderator. It actually was a chemistry question but considered unsalvageable because it was mainly foreign language and also a homework dump.

Andy

Thanks.

quartata

cc @Undo on 3 v 4

Andy

So...who wants to talk about ways to improve the terminology on metasmoke? I've got 3 questions based on what I've seen in the meta discussion and confusion we've had to clear up recently.

- What does True Postive / False Positive mean in metasmoke?
- How does a True Positive Autoflag differ from a True Positive SmokeDetector detection?
- How do we easily differentiate between the two?

quartata

> Thou shalt flag to 5, no more, no less. 5 shall be the number that thou shalt flag, and the number of the flagging shall be 5. 6 shalt thou not flag, neither flag thou 4, excepting that thou then proceed to 5. Seven is right out.

5

doppelgreener

6:51 PM

I think y'all would benefit from an infographic showing a bunch of posts arriving, being pulled out to smokedetector, being given to human reviewers which train algorithms, and those algorithms prompt when flags should get a few automatic flags to start with.

@quartata hahaha. :D

quartata

it's what the numbers say

John Dvorak

@doppelgreener I love that. Do you happen to have mad graphical skillz to donate by any chance?

DJMcMayhem

@quartata What an eccentric performance...

doppelgreener

@JohnDvorak i have good information layout skills but not the software for doing mad graphicz

DJMcMayhem

10/10

doppelgreener

6:53 PM

actually maybe GIMP will be enough

(i hate working with GIMP but it might do)

John Dvorak

Inkscape would be ideal

JAD

MS Paint, let's go

tripleee

@Andy not really where I can talk but good observations

doppelgreener

@JohnDvorak oooh. i'd forgotten about that one

right, i'm off to head home and have dinner. i'll check on my software options in an hour-ish.

DJMcMayhem

Hey now, you can actually make some really high quality stuff in MS paint

John Dvorak

6:55 PM

You can make a website in PHP as well.

WELZ

@DJMcMayhem They don't need one, they're good!

Savage

@Andy A "positive" result means that metasmoke recognized a post as spam. It showed "positive" indication that it matched the Regex we use to determine whether something is spam or not. "True" and "false" are markers of whether that "positive" indication was correct or not. "True" indicates that it is correct (the post is spam); "false" indicates that it is incorrect (the post is not spam).

John Dvorak

technically manual reports should be classified as false negatives rather than true positives, but I don't think Metasmoke splits the difference.

Catija

@JohnDvorak Yes.

If we were going to actually mention that half of the equation, we would want to discuss what a "negative" is.

John Dvorak

7:03 PM

I mean... cares about the difference

Andy

@Catija That answers from the detection perspective. We also have autoflagging true/false positives. I think the combination of the two is where we are confusing people

John Dvorak

we are indeed

Catija

@Andy Then maybe, I, too, am confused? If MS picking it up at all is a FP, then autoflagging is a FP... correct? So the confusing section is in the middle... posts that meet some of the MS regex requirements but not sufficient amount of them to actually trigger the autoflagging?

thesecretmaster

Just made a PR to add RSS.

SmokeDetector

[ metasmoke ] ci/circleci failure on 3a0d531: Your tests failed on CircleCI

Glorfindel

7:08 PM

justified/unjustified autoflags?

SmokeDetector

New metasmoke user 'Derte Trdelnik' created

Glorfindel

Alternatively, post a [single-word-request] question on English Language & Usage ...

SmokeDetector

[ SmokeDetector | MS ] Manually reported answer: Lists in VBScript by user9453049 on stackoverflow.com

tpu- by Glorfindel

thesecretmaster

Grrr... rails can't deal with RSSController

SmokeDetector

[ metasmoke ] ci/circleci failure on cb1d94f: Your tests failed on CircleCI

New metasmoke user 'oMiKeY' created

[ SmokeDetector | MS ] Bad keyword in link text in body: Can't Uncheck Windows 8 Target in Xamarin Project by CoderSue on stackoverflow.com

[ metasmoke ] ci/circleci success on cb59e93: Your tests passed on CircleCI!

CI on 6ff9588 succeeded.

thesecretmaster

7:25 PM

Or... I might have a different/better idea

I also added RSS output to /search

SmokeDetector

[ metasmoke ] ci/circleci success on 8af5625: Your tests passed on CircleCI!

thesecretmaster

Somebody merge me plz

Catija

7:40 PM

@Andy This is how I see it ^^^

John Dvorak

+1

You could also add a thin red line to the bottom half, pointing "false positives"

SmokeDetector

New metasmoke user 'Howdy_McGee' created

John Dvorak

getting progressively more impossible to see, ofc

Catija

@JohnDvorak The problem is that the FP/TP status is irrelevant of the autoflagging... It's just a side effect of the weighting. Trying to explain them all in one go is... not necessary?

doppelgreener

It's explaining mitochondria to a person who's curios how a tree works

"it grows up. It has leaves which make food from sunlight. It has roots which get food and water from the earth."

Catija

7:46 PM

That's crazy... you can't make food from sunlight!

Andy

I agree, the confusion I'm seeing, though, is that users go visit metasmoke and see something like this:

> The flags were 39 (86.67%) true positives (TPs) and 6 (13.33%) false positives (FPs).
The posts were 14 (87.5%) TPs and 2 (12.5%) FPs.

doppelgreener

That's true! I cheated! It actually just uses the sunlight to make useful food out what it already had.

@Andy Have you considered using different terminology on the dash?

Andy

That's what the discussion is about :)

1 hour ago, by Andy

- What does True Postive / False Positive mean in metasmoke?
- How does a True Positive Autoflag differ from a True Positive SmokeDetector detection?
- How do we easily differentiate between the two?

My goal is to figure out what we need to change to make this clearer

Especially to non-charcoal users that find a link to metasmoke

Catija

@Andy So, in that example, there were 45 total flags cast by autoflagging? So, divided by three makes 15 posts? (I can math)

doppelgreener

"39 flags were automatically cast on 14 posts which were verified as being actual spam and removed. 6 flags were automatically cast on 2 posts which were found to be non-spam, and the flags were reversed without post deletion. X posts were incorrectly removed as spam, and later found to not be spam."

Andy

7:52 PM

That wording makes sense. But, now we have TP/FP here: metasmoke.erwaysoftware.com/reasons

Those tp/fps mean something slightly different

doppelgreener

Also "X posts were destroyed as spam by accident, and later found to not be spam."

John Dvorak

we want to keep that at zero

doppelgreener

@JohnDvorak yep, agreed. Disclosing that it is 0 helps clarify what's going on.

Like the second sentence doesn't count posts we destroyed incorrectly. That's important to clarify. Because there is a category of improper removals, where does that fit in?

It fits in there in that sentence.

SmokeDetector

New metasmoke user 'Sam Woolerton' created

doppelgreener

@Andy where's the tp/fp language on that page?

Andy

7:59 PM

@doppelgreener Hover over the bars

doppelgreener

@Andy aha, will do when I'm back at a desktop PC. :)

SmokeDetector

[ SmokeDetector | MS ] Username similar to website in answer: Finding right configuration for a dedicated mariadb server - for huge innodb tables by Erce on stackoverflow.com

Catija

@Andy Yeah... that's why you don't use the same words to mean two different things.

I'm still confused how 14+2 and 15 are the same thing.

Mithrandir

@Catija wat

Catija

8:15 PM

@Mithrandir 39 + 6 = 45/3 = 15 Correct?

But this says 16 total posts (14+2):

28 mins ago, by Andy

> The flags were 39 (86.67%) true positives (TPs) and 6 (13.33%) false positives (FPs).
The posts were 14 (87.5%) TPs and 2 (12.5%) FPs.

The math is wrong somewhere.

SmokeDetector

fp- by Jake Symons

[ SmokeDetector | MS ] Link at end of answer: How show simple icons with only css by SpaceBlade387 on stackoverflow.com

New metasmoke user 'tfbninja' created

[ SmokeDetector | MS ] Offensive answer detected: I didn't get 4 outbreak packs for free by Ghost Gaming on gaming.SE

tpu- by Jake Symons

doppelgreener

8:32 PM

@SmokeDetector fp

SmokeDetector

@doppelgreener Registered answer as false positive and removed user from the blacklist.

Conflicting feedback on I didn't get 4 outbreak packs for free.

fp- by doppelgreener on How show simple icons with only css [MS]

doppelgreener

@SmokeDetector pretty standard unicode reference site

SmokeDetector

[ SmokeDetector | MS ] Few unique characters in body: stuffstuffstuffstuffstuffstuffstuffstuff by gkalikapersaud on stackoverflow.com

tp- by K.Dᴀᴠɪs

Undo

Let's do a thought experiment: Suppose we're right about spammers' incentives - if we get rid of the outliers that aren't deleted fast enough, they don't make money.

Also, suppose the spammers want to make money. That's pretty easy.

Now, suppose we were to turn autoflagging down to 1 or 2 on every site except one - say we move AU to 5.

If spammers are incented to make money, and money is made with outliers, and suddenly AU has no outliers... we should see spam volume fall on AU and possibly raise on other sites.

M.A.R.

meta.stackexchange.com/a/307655/280934 -- no kidding, TeX is like Disneyland

Undo

8:41 PM

So maybe an opt-in scenario would work with the right management - we'd essentially chase the spammers around the network, new sites would bear the burden of the spam wave, ask for us to help... and, eventually, spam volume across the network would drop off a cliff.

Nice thing about this is it's very easy to test. We'd kill the 5-flag experiment, turn one site up to 5, see what happens

SmokeDetector

[ SmokeDetector | MS ] Offensive answer detected: Failed - Network error downloading my Google Takeout by hard gore on webapps.SE

M.A.R.

@SmokeDetector K

@Undo you sound too maybe to make a decision

Undo

?

M.A.R.

@SmokeDetector WTH is up with that thread?

Mithrandir

@SmokeDetector tpu-

WELZ

8:46 PM

@M.A.R. IKR?!

M.A.R.

@Undo have you decided yet how to do this?

Undo

What do you mean? I'm not going to just do it on my own.

Mithrandir

@SmokeDetector n

SmokeDetector

Conflicting feedback on I didn't get 4 outbreak packs for free.

Mithrandir

@Undo Ehh...

WELZ

8:48 PM

@SmokeDetector @doppelgreener fp?

Undo

It's pretty straightforward, though. Make change, collect data, observe data. I bet we could convince AU or Drupal to take five flags for a few weeks.

WELZ

naa at the best.

Mithrandir

(or Apple.SE since we have a resident Apple.SE mod)

Undo

That'd work too

Mithrandir

I'm not a fan of lowering the other sites though.

Undo

8:50 PM

@Mithrandir Lowering is optional - it'd just be to confirm a cause-effect relationship.

Mithrandir

Also, I don't get where that conclusion came from, that it'll suddenly plummet.

doppelgreener

@WELZ it's at least not tp anything

@SmokeDetector naa-

(there we go)

Undo

@Mithrandir It's a theory: If spammers want to make money, and we make it harder for them to make money on one site, they'll probably go somewhere else eventually.

It's a way to confirm that they rely on the outliers, without convincing the whole network to take 5 flags.

Mithrandir

They already are other places as well. I stalk our pharma spammers on Google.

Undo

Indeed.

WELZ

8:51 PM

@Undo We have 2 GD mods that are involved with Charcoal. (Cai and Wrzlprmft) they may be on board with this as well.

Catija

@Undo Would they change their target sites based on which sites were failing to remove spam quickly?

WELZ

Unlikely that they notice.

Undo

@Catija That would be the theory, yes.

@WELZ If their income is in some way tied to TTD, they'd notice

Mithrandir

(Which makes me wonder about setting up Smokey on other sites they're hitting and truly be spam crusaders of the internet)

Catija

MSE is probably an easy OK, considering..

WELZ

8:52 PM

@Undo I (personally) don't think it is.

Seth

What's the new question?

WELZ

I We should reach out to one of those agencies and figure out how it works

Undo

@Seth Trying to figure out what spammers' incentive is. Theory is that it's tied to TTD - I'm trying to see how we could confirm that.

John Dvorak

Try randomizing the TDD and seeing the spammer income? :P

Catija

Honey pots...

JAD

8:54 PM

@JohnDvorak I'm not sure they publish their revenue numbers publicly

Seth

@Undo I have my doubts about that, but science is always fun.

@JAD Maybe if we found their tax returns :P

Undo

Basically, force one site to have a lower TTD than others and see what happens.

Seth

not tax returns, but you know what I meant.. joke ruined >.>

WELZ

it has to also be a higher volume site (not too high though)

Undo

AD would do quite nicely

SmokeDetector

8:57 PM

New metasmoke user 'Jacob G.' created

WELZ

Just btw, TTD really does matter (regarding Search Engines) searching these spam sites on Google will bring up other forums. (when you go there it shows deleted) but the search engines have already picked it up

Undo

Expectation would be less spam on a site as TTD goes down. If that migrates somewhere else on the network, we can chase it - if it doesn't, awesome.

SmokeDetector

[ SmokeDetector | MS ] Few unique characters in answer, repeating words in answer: Did Kakashi ever love Rin or at least feel anything for her? by user38480 on anime.SE

naa- by Glorfindel

Undo

And if it isn't affected, we ask what the heck their incentives actually are.

Seth

@Undo I'm really not convinced the their systems are that smart.

Undo

8:59 PM

Perhaps not - but ultimately, someone is paying them. They might be smarter.

You could even set up a bigger experiment - squeeze one site, then another, then another. Make a nice little bar graph animation.

Seth

I doubt they have a metaspark where they track how successful their spam is and find new targets :P

Transcript for