Charcoal HQ

tripleee

12:00

or just try it until it possibly starts working?

SmokeDetector

[ SmokeDetector | MS ] Few unique characters in answer, offensive answer detected, repeating characters in answer: Is there a way to clear your printed text in python? by LOL U GOT TROLLED on stackoverflow.com

tripleee

@SmokeDetector k

ArtOfCode

the revision thing in the top-right corner of metasmoke tells you what revision is currently deployed

tripleee

@ArtOfCode so when it's no longer showing 7467ac6 give it a go?

ArtOfCode

basically, yeah

SmokeDetector

12:09

[ metasmoke ] ci/circleci failure on 40d0756: Your tests failed on CircleCI

tripleee

so the query format is like /api/posts/search/regex?query=foo%3Cbar or what?

ArtOfCode

@tripleee yep, query is the regex you want to use

and a key, obviously

tripleee

yeah, thanks

SmokeDetector

12:27

[ metasmoke ] ci/circleci success on a763523: Your tests passed on CircleCI!

Mithrandir

@trip how do you look up who owns a domain?

I'm trying to look up who owns underwise.com but I can't figure out how.

thesecretmaster

whois search?

Domain Name: underwise.com
Registrar URL: godaddy.com
Registrant Name: tian fei
Registrant Organization:
Name Server: NS63.DOMAINCONTROL.COM
Name Server: NS64.DOMAINCONTROL.COM
DNSSEC: unsigned

tripleee

@thesecretmaster yeah, it's just that godaddy's whois is apparently "updated" so you don't get results from services like who.is or even your regular whois command-line client

directly

@thesecretmaster that only reveals that it's available from godaddy

thesecretmaster

And a name, right?

tripleee

@thesecretmaster well yeah, but that's often forged and/or uninformative

thesecretmaster

12:31

Aw

tripleee

@Mithrandir pastebin.com/cEWRsdsx

Mithrandir

that name looks like a Chinese sea goddes, so yeah, not helpful

@tripleee Thanks.

(background: they're taking like everything on their site from SE without attribution)

thesecretmaster

@tripleee How'd you get that? I tried using the -h flag on my command line (arch), and didn't get anything.

Mithrandir

(Even Meta)

ArtOfCode

@thesecretmaster -h whois.godaddy.com

you need the WHOIS server address as well as the flag

thesecretmaster

12:33

Yeah, I just tried that.

ArtOfCode

I'm gonna guess the arch client is different then :)

tripleee

not familiar with Arch but if all else fails, direct telnet to port 43

there are multiple generations of the whois protocol, the referral logic was a big change but it looks like godaddy (unilaterally?) changed something on their side recently

FWIW my result is from MacOS Sierra

thesecretmaster

I think the arch client is lying to me: -h HOST, --host HOST connect to server HOST

ArtOfCode

@tripleee mine from Ubuntu

tripleee

if it's using that as the referral server, not the final destination, that would be a problem, but this is pure speculation

strace would probably help you figure out where it's going wrong if you really want to know

ArtOfCode

12:36

can confirm that $ telnet whois.godaddy.com 43 works, though

thesecretmaster

Eh, I don't care enough. I don't use whois very often

Mithrandir

This makes me sad. They've taken it without even keeping the formatting. If you're going to plagiarize, do it well!

thesecretmaster

They don't seem very wise...

tripleee

@ArtOfCode some servers use the input as a wildcard, you have to figure out their protocol for submitting a verbatim query, IIRC it's =query or something like that

J F

> several seconds ago

lol

ArtOfCode

12:37

@tripleee You can just type the domain name with GD

tripleee

@ArtOfCode yeah, thankfully

Magisch

Hey people

Mithrandir

yo

Magisch

My arms feel like they're made of butter. I've been carrying boxes all morning for the moving of part of the office

Mithrandir

get some bread

Magisch

12:42

But bread is like all carbs

like 90% carbs

Mithrandir

but it'll be good with the butter

Henders

I think it was a 'bread and butter' joke...

groans

Magisch

Maybe I was being dense on purpose to ruin the joke :D

Henders

@Magisch I don't think there was any knead for that...

tripleee

@Henders aaaaargh

Magisch

12:44

wow

that was so bad it was actually funny again

Mithrandir

the level of punniness in here is rising

Henders

@Mithrandir Just like a freshly made bloomer?

ArtOfCode

12:58

Autoflagging RFC: #AF4. What do we need to expand autoflagging further?

9

your comments please

SmokeDetector

[ SmokeDetector | MS ] Few unique characters in answer: Uniform Integrability of exponential random variable by phdmba7of12 on math.SE

tp- by ArtOfCode

Mithrandir

sd - v

ArtOfCode

@quartata you were talking about security for autoflagging expansion a few days ago; there's an RFC you can chuck comments/suggestions on there

SmokeDetector

[ SmokeDetector | MS ] Bad keyword in body, blacklisted website in body, link at end of body, repeated URL at end of long post: You need a consume; stick your by user725486 on askubuntu.com

tpu- by Ashish Ahuja

[ SmokeDetector | MS ] Bad keyword with email in body: Please answer this question, Can you install elementary os on an ssd hard drive? by Chris McCabe on elementaryos.SE

fp- by Mithrandir

Henders

13:26

!!/location

SmokeDetector

@Henders ArtOfCode/EC2

[ SmokeDetector | MS ] URL in title, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: puresupplements786.com/stackt-360/ by lucy fennesey on astronomy.SE

tpu- by Henders

[ SmokeDetector | MS ] URL in title, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: puresupplements786.com/stackt-360/ by user251649 on apple.SE

tpu- by Ferrybig

[ SmokeDetector | MS ] Blacklisted website in body: How to Set up review page at word press by Sarah on stackoverflow.com

tpu- by Mithrandir

API quota rolled over with 8933 requests remaining. Current quota: 19999.

stackoverflow: 377
math: 337
superuser: 76
askubuntu: 67
physics: 61
serverfault: 52
codegolf: 48
english: 43
apple: 40
travel: 27
workplace: 23
unix: 23
electronics: 21
security: 21
gis: 20
drupal: 19
movies: 19
ell: 18
mathoverflow.net: 18
scifi: 17
dba: 17
salesforce: 17
wordpress: 15
ru.stackoverflow: 15
meta: 15
stats: 15
sharepoint: 15
arduino: 13
gaming: 13
magento: 13
android: 13
academia: 12
interpersonal: 11
diy: 11
graphicdesign: 11
softwareengineering: 10
politics: 10
webapps: 9
islam: 9

SmokeDetector

13:57

[ SmokeDetector | MS ] Username similar to website in answer: What is a non invasive, large, structural species of Bamboo? by Lewis Bamboo on gardening.SE

Mithrandir

sd f

for now

Undo

RFC identifiers are becoming increasingly complex :P

Andy

@ArtOfCode I'm going to comment here because I think it needs a bit of discussion. Based on the responses to bluefeet's post about my bot (here) for reference, I don't think the community is going to support us having the ability to unilaterally nuke anything.

SmokeDetector

[ SmokeDetector | MS ] Username similar to website in answer: Claiming my own country by Guanduania on law.SE

Andy

I have the accuracy you are striving for and it's about comments, and the community still doesn't want it done automatically - unless it's something that SE runs and maintains

SmokeDetector

14:02

tpu- by Mithrandir

ArtOfCode

@Andy I think you're right... which is kinda why I was going to propose us discussing it, then taking it to Pops and surreptitiously updating our meta post after we've already been doing it for a while :)

Undo

it'd probably be good to at least have a show of being open about it

That said, I tend to think that if we go in with a mountain of data (which we have) showing a bunch of nines (or maybe some zeros?), it'd be pretty easy to shoot down the principled arguments

Or I'd hope so. Also, this time we could probably start on mSE :P

Andy

@Undo It hasn't helped in my case.

@Undo You don't want to have that fight again? :)

Undo

egads

ArtOfCode

well we're not about to post on every single meta site, sooo...

Undo

14:05

That one blindsided me. Like playing football and someone hits you in the face with a frisbee.

Andy

lol

Undo

@Andy True, although your case is slightly different - we've got a lot more visibility and public process built up around everything

I've got reservations about running the comment flagger without oversight too. We've got oversight galore.

Andy

I'd love to be wrong, but I think we'd hit the same community split - yup, automate all the things vs. nope, human brain required

Undo

It's a Big Deal when we autoflag a false positive, and we're really good at getting feedback on everything quickly. You could even do something like require feedback before actually killing it.

ultimately, I know you're right and I really want you to be wrong

ArtOfCode

That could work. If we went for 5 flags (with high enough accuracy, like 99.99%), then required feedback for the 6th...actually, that'd create a nice race condition between FIRE and MS

which I suspect MS would win

Undo

14:08

4, then feedback

One human involved to destroy something a trusted instance thinks is spam. Pretty safe.

Of course, we all live in this system and know how resilient it is. Other folks see "you want a robot to delete my stuff"

ArtOfCode

Okay, let's think about trusted instances. I like that idea and I don't like it at the same time.

I like it because oversight and code review and obvious accountability

but I don't because duplication and slower deploy cycles

Undo

That said, I don't have an issue with going to 4 without making a big deal out of it - it's a natural progression and doesn't change the dynamic much. 5 or 6 is different.

@ArtOfCode Don't know about the deploy cycles. It'd still be as fast as it is now to deploy a blacklist item to public and get 3 flags on it, and if something needs to be un-blacklisted removing it from the public instance would keep it from being flagged

(I'm seeing a system where the public instance reports to metasmoke, trusted one confirms)

Some timing things to work out there, but not too bad

ArtOfCode

@Undo Aye, you could get 3 flags on a post from the public instance, but you can't get a website blacklisted and eligible for 6 flags so quickly - that has to wait for the weekly review-and-merge

Undo

You'd just need to wait until the deploy cycle to jump from 3 to 4-6 flags. It wouldn't be too bad.

@ArtOfCode that

Which might be a good thing anyway

Andy

Can someone briefly explain what the trusted instances will be doing?

Undo

14:15

@Andy TL;DR: Run a headless SmokeDetector instance on a machine we trust, at a Git revision we trust, then require it to detect a post before we allow more than 3 autoflags.

To get around the whole "remote code execution as a service" thing

afk for a minute

Andy

Ah. Ok. That kind of makes sense. My issue with that is that our blacklists and such are in git too.

ArtOfCode

@Andy what issue does that create?

Also, rewinding a bit... I think opposition actually might not be as bad as we're dreading here. Our meta post had almost zero opposition - I think I remember one person seriously objecting), and that was after we already started doing it.

Undo

That's why we have two - one to keep our agility, the other to keep integrity.

@ArtOfCode aside from the frisbee lobbing, yeah

ArtOfCode

Now, we might get more opposition to 6 flags, but certainly 4 and possibly 5 might not meet too much more opposition

@Undo yeah, I was ignoring TL for now :)

Undo

@ArtOfCode I think we did a pretty smart thing there by allowing people to sign up and reap rewards from it. Made folks feel like they were part of it (hey, free flags) and thus less likely to dislike it

ArtOfCode

14:19

Also true

Andy

@Undo Does it get you that integrity though? I think I'm coming back to "blacklists shouldn't be in the code" argument. If the majority of our code updates are actually blacklist/watchlist updates, when the trusted instances are updated will each of those blacklists actually be checked? Or will we just say it's time to update? If we are just doing this to watch the code of smokey, I think it's a great idea.

Henders

4 mins ago, by Undo

To get around the whole "remote code execution as a service" thing

What does this mean?

Undo

@Andy I was thinking we'd review everything, including blacklists.

ArtOfCode

@Henders anyone with git privs, and to a lesser extent any code admin, can remotely execute code using Smokey on someone else's machine

Undo

@Henders !!/pull allows anyone with GH privs to immediately run anything they want. Which is terrifying

ArtOfCode

14:20

Git privs: commit some malicious code, !!/pull, and your code is live on the current host

Undo

So far, pallet-of-bricks has been a good deterrent. Having honest people helps too.

Henders

Right, gotcha

Andy

@ArtOfCode I had very little opposition to my announcement of what I was doing too. There is a difference between "Look what I can do! Someone needs to click that button now and confirm all the work I just made for them" and "Well...let's not make them click that button any more" apparently.

ArtOfCode

@Andy I think the idea is that we have a weekly (or other time-period-based) check where the trusted hosts check the entire diff from the last week, and merge only if they think it's good

Undo

This diff isn't too hard to read at all

Watched sites can be ignored there

@Andy This is also true. We do have more grey area between "bunch of humans involved" and "no humans involved" - might have to exploit that.

Ease folks into it, in tiny steps. 4, to 5, to 6 with human oversight, to straight 6.

(meaning 'at least 1 feedback')

Mithrandir

14:25

...I don't think I'd be comfortable with a 6 unless the weight was 1,200 or so.

ArtOfCode

@Mithrandir What's your reasoning behind that?

Consider that 280 weight is 99.99% accurate

Andy

@Undo Agreed. So, now my question is why should the trusted instances be a week behind? What does that really get us?

ArtOfCode

@Andy trusted people not having to read tiny diffs a hundred times a day

Andy

(Um...I realized that may have come off as hostile, sorry)

Undo

It'd be good to keep an aura of importance around those reviews - if they're every day or two, we may risk people becoming complacent. Every week or two might keep it feeling like a Big Deal

Not at all sold on that, just thinking out loud

ArtOfCode

14:28

there'd be a very small number of those people - our current situation is okay, because there are plenty of people to split diff-reading between, but when you take it down to two or three and require each of them to independently read every diff and judge it... that's a lotta dedication

Mithrandir

@ArtOfCode It's irreversible without a mod of that site. If there's a singe FP that slips through to that, the OP suffers a -100 rep penalty, and an IP block, and other assorted consequences. That 1 FP is a big deal. Spam flags are one of the most dangerous things on SE - they're almost the only way a few users can unilaterally delete a post and give other consequences without moderator oversight.

Undo

No reason everyone can't be invited to take a look, too

ArtOfCode

@Undo Sure, but your trusted people have to independently make a call on it

@Mithrandir I dug some data out on this a while back. Lemme see if I can find it.

Henders

Is it necessary for anyone with privilege to be able to !!/pull? Couldn't it be like a code_admin command insteaD?

Undo

In theory, this'd be a sanity check for "did someone blacklist spaces". Anyone throwing a red flag would probably shut it down while we discuss stuff.

@Henders Pulling is useless without ability to change code

I mean, it might be useful for irritating us, but not much more

Henders

14:31

Oh that's true, I didn't think about that...

ArtOfCode

Hang on. Metasmoke has a diff view for its own code. We could probably build a tool around that - let metasmoke collect the diff each week, and get trusted people to approve/reject/comment it, like PR reviews

A reject raises Big Red Warnings, tm.

Undo

Just automatically post a pull request on GitHub

ArtOfCode

Can't find my data... Anyway, @Mith, that data said that something along the lines of 95% of people whose posts got spam-nuked didn't get any penalty because they were already 1-rep; then something like 50% of the remaining 5% didn't get the full penalty because they were sub-100 rep.

@Undo time to resurrect GH required reviews and metasmoke CI?

Mithrandir

@ArtOfCode That's irrelevant, though. If a single person gets the -100 when they didn't deserve it, that's too much, MO. Also, I don't want to have those sub-100 rep users lose all their rep either.

Andy

Someone impress me with their Metasmoke search capabilities: Which posts have we autoflagged as spam that are not spam?

(That'll be a question asked on any meta post we make)

ArtOfCode

14:37

@Andy metasmoke.erwaysoftware.com/flagging/logs?filter=fps

we have a view dedicated to it :)

Andy

That's just cheating :P

ArtOfCode

5 in the last month

Mithrandir

32 altogether, looks like

ArtOfCode

@Mith that seems extreme to me. Every spam detection system has false positives; yes, these consequences are pretty severe, but they're also easily reversed by a mod. I don't think one accident in ten thousand - or even in more, if we wanted, is too great a price to pay.

Mithrandir

@ArtOfCode well, I may be being a bit too paranoid. But a single FP will make people lose some faith in us.

Henders

14:40

"I got -100 and IP blocked because of smoke detector"

Not a great headline :P

ArtOfCode

Some, yes. Our job is to accept that, and offset it by showing people what we can give them in return for that one false positive.

Andy

@ArtOfCode Good read some of the comments on bluefeet's MSO post about the comment flagger. Users are attached to their content. Deleting a comment is contentious and has no penalty on the poster. Spam deleting a valid post is very different from a comment. That's what the user's will see. They won't see your numbers.

ArtOfCode

@Henders "Yup, seen it, system dun goofed. Penalties reversed, and we're working on making the system better."

Andy

@ArtOfCode Repeat 5 times a month though.

Henders

Why not work on making the system better first? Are we in a rush to get to a full-smokey-nuke stage?

ArtOfCode

14:42

@Andy You're right, but I wonder if there might be some room for distinction between users' own content and spammers' content, which nobody likes.

Undo

Remember: You can't get a 100 rep penalty if you've got 1 rep to start with

ArtOfCode

@Andy less than that, because we'll require higher accuracy

Andy

Ah. True

ArtOfCode

could also lock it to 1 rep to avoid the rep penalty

Undo

280/1/1 gives us 99.99% 25117 tp v. 2 fp.

Andy

14:43

@ArtOfCode Sure there is, but we are talking about false positives here. It's valid content that we thought was spam.

ArtOfCode

@Henders We will, but it'll be continuous improvement. If we kept trying to make it better before launch, beyond a certain point we'd just never launch.

@Undo did we not have the same 2 fp when we started in January? So literally zero FPs actually autoflagged with those settings?

Undo

probably

Sconibulus

@Andy Deleting comments is also seen as low-value by many users, I don't think Spam is seen in the same light

Andy

The solution, obviously, is to fly someone out to the homes of those posting false positives and yell at them so they learn not to post spam like content. I volunteer for these paid trips.

ArtOfCode

I like it

Undo

14:44

350/1/1 gives us 17.2k true positives, zero false positives. Hard to argue with that

Andy

@Undo Can I pick out a handful of users that will? :)

ArtOfCode

@Andy Aye. I'm just wondering if that's not slightly different to comments, which are always user content, as opposed to spam/spam FPs which are only sometimes user content?

Undo

Probably :P

ArtOfCode

@Undo Bet you can drop that to 300 and get pretty similar stats

Undo

one or two at 300. I tried :P

Andy

14:46

@ArtOfCode I'm not completely understanding.

Undo

He's saying that folks might not be attached to spam like they are with something they posted in good faith

ArtOfCode

@Andy With comments, a TP is user content. So is an FP. With spam, an FP is still user content, but a TP is not - it's trash.

Andy

Ah. My lightbulb has flickered.

ArtOfCode

310 weight for 1 FP

it's not even an FP, it's NAA

and absolutely terrible while it's at it

Undo

Hackles go up when you delete people's stuff. If you can make it clear that "if you're reading this, this will probably never touch you" maybe they'll de-hackle.

that... almost looks like spam

ArtOfCode

14:48

aye

Undo

yeah, that's spam. Deleted as such

And user destroyed.

quartata

What's going on here?

Undo

Autoflagging discussion

ArtOfCode

Heh, that was my feedback. Changing to TP.

SmokeDetector

naa by ArtOfCode on Zend e-commerce extension [MS]

quartata

14:48

Is this about autoflagging

SmokeDetector

tpu by ArtOfCode on Zend e-commerce extension [MS]

ArtOfCode

This is the other FP at 280 weight, which is also terrible

probably doesn't quite count as spam because Patents is weird, but it's not good content

Mithrandir

@ArtOfCode that looks... very, very, similar to the other one.

Undo

Looks like foreign chars from before we supported them

quartata

So we're considering increasing the amount of flags based on weight?

Undo

14:50

Yep

ArtOfCode

@quartata or some other metric, yeah

Undo

And how to easy folks into it

quartata

Hmmm

ArtOfCode

301 weight gets you 100% accuracy

Undo

lol

ArtOfCode

14:51

and 17462 TP

quartata

Presumably we don't want to use 6 flags

ArtOfCode

you're still looking at 30-40/day there

Undo

@quartata For now

Ultimately, the end goal has always been to delete spam. 6 flags is the way to do that, eventually

Also, this:

Mar 18 '16 at 19:33, by Undo

observation: projects stop growing when they become change-averse.

ArtOfCode

heh, the last part of this isn't quite so true any more

Andy

I think our biggest problem is going to be handling the community reaction. The technical side we've got figured out. It's just changing a few numbers and letting the system do it's thing.

quartata

14:55

Not exactly

ArtOfCode

it's a few people writing a few more bits of code, but it's not difficult

Andy

Pedants :P

Undo

code is technically numbers

ArtOfCode

numbers are technically electricity

quartata

My first thought was assigning a new threshold for each level of flags beyond the 98.5%/3 flags

ArtOfCode

14:58

99.5%

Undo

Yep, that's what I was thinking

quartata

But it doesn't fit with how we do criterion. Do we use 4 flags just because one user has a criterion that fits?

Undo

Check the post weight against the accuracy numbers we have.

quartata

The problen is that users have a criterion that determines accuracy

Undo

Basically do the same thing flag condition preview does

The criterion determines the minimum accuracy the user wants. Doesn't mean we can't be more confident about a single post

Take the post weight, rep, etc. and run it against previous posts to see what our track record with that post profile is. There's your accuracy number

ArtOfCode

15:00

if post.accuracy >= criterion.accuracy
  post.spam_flag

Undo

Then map that to a max flag count using whatever magic you want

quartata

Hmmmmm

ArtOfCode

actually, it'd make more sense to do it the other way around

Undo

@ArtOfCode More (post.accuracy * n).to_i.times do |i| { p.flag } or similar

We just set the number of times we're willing to flag.

afk, should get something done today. Cya folks :)

Henders

@Undo I'm going to get that on a T-Shirt. Although... I like my current T-Shirt so maybe I'll just stick with this one.

Andy

15:03

You should all be pleased to know that I horribly broke a system...and it wasn't Smokey.

You're welcome

ArtOfCode

standing ovation

J F

@SmokeDetector k

quartata

In that case it no longer makes sense to flag based on available criterion for the 3 flag level and only use accuracy to determine extra flags

In effect what we're suggestinv really is that the system should maintain its own minimum criterion, and users canbe more conservative if they want to be excluded sometimes

this has always kind of implicitly existed

Sconibulus

Oh, so the criterion we set don't matter much?

ArtOfCode

yes they do

they determine whether or not your flag gets used

they are quite literally one of the central tenets of the system

what quartata is describing is already how it works, but only in the fact that your conditions have to be at least 99.5% accurate - that 99.5% is the system-minimum accuracy

quartata

15:08

Yeah Art gets it

Sconibulus

right, but I mean, I decided to bump up the required minimum weight in exchange for having a higher maximum rep, in case someone who was borderline at first got worse; if there is a minimum threshold for any flags to be cast, then that kind of tradeoff doesn't work the way I thought it did

quartata

But

From a code standpoint it's beneficial to maintain multiple sets of minimum criterion and check posts against that rather than "searching for available flaggers, then try to get more if we meet a certain accuracy threshold"

ArtOfCode

@Sconibulus When you're setting up your conditions, the percentage at the bottom that you get from hitting Preview is the ultimate governor of when your flags get used

if it's < 99.5%, it'll be rejected

above that, the higher it is the less your flags get used (but the more accurate it is when they do get used)

Sconibulus

oh, huh

so the actual conditions don't matter much?

ArtOfCode

yes they do

quartata

15:14

Bonus points for if we make these new criterion visible so that people don't have to experiment until they get the lowest possible

@Sconibulus currently the system only autoflags if there's someone with the right criterion available

with no consideration for accuracy outside of when the criterion were originally set

now that accuracy matters at flagging time this doesn't make sense, hence my wall of text

ArtOfCode

We can still use the same condition model, we're just moving from matching the numbers to matching the accuracy

bassxzero

15:31

What command line email tool should I use for Linux servers? I'm looking for something that is widely available across the Linux distros.

John Dvorak

SSH to Gmail :-P

I mean, IMAP

quartata

mutt if you want something interactive, mailx if you want to use it from scripts

bassxzero

mailx it is then. Thank you

quartata

@ArtOfCode elaborate?

SmokeDetector

[ SmokeDetector | MS ] Few unique characters in answer: Best way to send layouts with editable text to writers/editors who don't have design software by testbot on graphicdesign.SE

tpu- by Henders

quartata

15:44

if the concern is finding the actual criterion that's not hard

ArtOfCode

We keep the model we have, but instead of being a query for where weight, rep and reasons count are above the required, it's a query for where accuracy is greater than required

Henders

Accuracy of the user?

SmokeDetector

[ SmokeDetector | MS ] Username similar to website in answer: djago-rest-auth multi token by user by Raja Prasanna B on stackoverflow.com

naa- by J F

quartata

But that's accuracy of the user

And what are "required" in the first query

we don't have the minimum criterion as requirements that's the point

SmokeDetector

[ SmokeDetector | MS ] Username similar to website in answer: Multiple Token Authentication in Django Rest Framework by Raja Prasanna B on stackoverflow.com

Henders

15:54

Second time... Spam?

SmokeDetector

tpu- by J F

tpu- by Mithrandir on djago-rest-auth multi token by user [MS]

quartata

Here one second let me get off of mobile and then I can try to present my case a second time

This time with proper spelling

J F

> Did you mean "speling"?

quartata

speeeling

OK here's the scoop

Henders

. omlet.co.uk/images/cache/1000/660/Eton_aluminium_food_scoop.jpg

quartata

16:00

Currently when a user signs up for autoflagging, they put in whatever criterion they want and the system verifies that it has 99.5% accuracy before adding it

Then when a post is reported, Metasmoke checks to see if there are any users with criterion that match the posts, up to 3

ArtOfCode

@quartata So currently, when we get a post, we do a query that goes something like

FlagCondition.where('min_weight <= #{post.weight} AND reason_count <= #{post.reason_count} AND user_rep >= '#{post.user_rep}')

quartata

When Metasmoke does this, accuracy isn't involved at all, it just implicitly knows that these criterion have to be good enough otherwise they wouldn't be here

ArtOfCode

In a new system, we'd be doing a query more like FlagCondition.where('accuracy <= #{post.accuracy}')

quartata

But that doesn't make sense

ArtOfCode

Oh yes it does :P

why does it not?

quartata

16:03

First off what is post.accuracy

ArtOfCode

a theoretical calculated value based on the post's reason count, weight, and user rep

We take those three, and find out how accurate that combination of values has been in the past

SmokeDetector

Restart: API quota is 18424.

quartata

OK

But then you're not actually enforcing the users' criterion

You have to do both

Calculate accuracy for how many flags and compare criterion for which users to pick

That's ugly

ArtOfCode

there you go, @quartata

Andy

smashes reject button without looking

Just because it was red...and had "Do not push" on it

J F

16:15

Huh?

quartata

@ArtOfCode Do you get my point?

If not I can construct an example

Andy

An example would be helpful

ArtOfCode

speaking of, my example didn't actually make much sense... lemme try again

Andy

Ah, well, my button smashing was correct then :)

quartata

Suppose a post is 180/2/1. Someone with 150/3/1 could flag it even though the number of reasons don't match

You have to check both accuracy and criterion

What I have in mind takes accuracy completely out of the picture except for a once a day cron job

Calculating accuracy every post is expensive anyways

ArtOfCode

16:25

@quartata what do you have in mind?

quartata

calculatinv the criterion that maximizes TPs caught for each accuracy threshold

that becomes the minimum

ArtOfCode

there you go, this makes more sense @quartata

Andy

some_formula_involving_accuracy - heh

Serious question, aren't the >= and <= backwards? (Or, maybe I don't know ruby well enough)

ArtOfCode

No, it's just counter-intuitive

(that's SQL, anyway ;) )

A flag condition has a minimum reason count, therefore the post reason count has to be greater than that value

post_reason_count > minimum_count

minimum_count < post_reason_count

Andy

Ah. Makes sense

wolfboyft

16:52

What is the actual purpose of this channel?

You can't really make diamonds here, can you?

Is it a general chat room or what

rene

spam, spam and spam

and diamonds

wolfboyft

So it's general.

but spam is often used

rene

well, there is a link, right? charcoal-se.org

wolfboyft

ohh

So we just spam so that the bot knows what spam is?

or what

Andy

No. The bot in this room watches the network for new posts and determines if they are spam or not. If it is spam, it is posted here and elsewhere so that users can handle it

wolfboyft

16:55

right...

¯_(ツ)_/¯

ciao!

oups one sec ¯\_(ツ)_/¯

there we go

rene

awesome

wolfboyft

Discord actually has a command for that emoticon (it's quite famous.) /shrug <text> = <text> ¯\_(ツ)_/¯

Byte Commander

17:13

No such luxury here.

SmokeDetector

17:25

[ SmokeDetector | MS ] Offensive answer detected: Why is hibernate returning a proxy object? by Fuzz Nuts on stackoverflow.com

tpu- by Henders

SmokeDetector

17:44

[ SmokeDetector | MS ] Bad keyword in body: How to access JSON data in classic ASP by Ted Jang on stackoverflow.com

fp- by J F

J F

sd - why

SmokeDetector

1. [:39429731] <skipped>
2. [:39429721] Body - Position 2063-2069: Sbobet, Position 2072-2078: Sbobet

J F

15 hours ago, by tripleee

!!/blacklist-keyword sbobet

@tripleee why

DavidPostill

@wolfboyft No, but we have some regulars here that became moderators, hence "where diamonds are made" :)

Does anybody know how many?

Glorfindel

Jul 17 at 11:42, by Magisch

Plus there's like what 10 moderators in this project.

Some of them (but not many) were moderators already.

DavidPostill

17:49

I suppose it's not surprising as mods hate spam ;)

Transcript for