Charcoal HQ

12:01 AM

@ArtOfCode Figure out who's gonna post it, maybe find someone who's colorblind and ask again if that graph is impossible to read.

ArtOfCode

@Undo done the latter, ACM confirmed it is

Undo

Awesome

ArtOfCode

(readable, that is, not impossible)

Undo

I think we can agree this time to do mSE :P

ArtOfCode

Yeah, let's... not... have that one again

thesecretmaster

12:02 AM

@ArtOfCode [tag:status-completed]

Undo

Do you want to change 'the settings we're proposing for 5 flags' to 'considering'? Don't want to get locked in the 'meta approved exactly this' box.

ArtOfCode

@Undo works for me

Undo

Question whether we should strike the 'sites with lower deletion thresholds' section for brevity - if the question is raised it's a simple one to answer.

ArtOfCode

that also makes sense

It's pretty long, and about to get longer

Undo

Also @thesecretmaster there's still something weird about flag false positives on SO dash; compare to search.

It's claiming 22 false positives over three months opposed to ~22 all time - is that on a flags or posts basis?

thesecretmaster

12:09 AM

I thought @ArtOfCode fixed it, lemme look

ArtOfCode

might not have extended the fix to FPs

thesecretmaster

Looks like you did

commit

@Undo That's flags -- top line is flags, bottom is posts

See "the flags were" vs. "the posts were"

Undo

That works. Wanted to make sure that's what it is.

SmokeDetector

API quota rolled over with 10601 requests remaining. Current quota: 19999.

math: 260
stackoverflow: 135
superuser: 49
physics: 37
english: 26
codegolf: 25
askubuntu: 24
apple: 19
electronics: 19
scifi: 17
travel: 16
serverfault: 13
unix: 12
graphicdesign: 11
wordpress: 10
academia: 10
bitcoin: 10
android: 9
webapps: 9
security: 8
diy: 8
mathoverflow.net: 8
quant: 7
ethereum: 7
ell: 7
tex: 7
history: 7
mechanics: 7
es.stackoverflow: 7
gaming: 7
rpg: 6
arduino: 6
drupal: 6
judaism: 6
dba: 6
gis: 6
softwareengineering: 5
workplace: 5
mythology: 5
raspberrypi: 5
mathematica: 5

Undo

@SmokeDetector Does anyone do anything with these messages any more?

thesecretmaster

12:16 AM

Found a bug: Dividing by 0 produces NaN

Undo

more 'math' than 'bug' ;P

thesecretmaster

I don't know how to handle that case though. The percentage really is undefined.

ArtOfCode

eh, not sure it really needs handling

So who do we want to post this meta post?

Andy seems like an obvious choice for continutity

SmokeDetector

tpu by ArtOfCode

fp by ArtOfCode on Hash map without collision check [MS]

quartata

@Undo one of the plans I had for the original NG was a more comprehensive (and less huge) end of day report

let me find my notes

thesecretmaster

12:28 AM

@ArtOfCode Just looked at the meta again -- I feel like it's missing actual stats on expected decrease in ttd. Am I missing something?

ArtOfCode

@thesecretmaster Yes, they're coming

thesecretmaster

Ah, OK

ArtOfCode

waiting for today's metasmoke dump, then I can grab the stats from the experiment

thesecretmaster

It was looking like 5 flags was ~30 seconds better than 3, right?

ArtOfCode

20

roughly a 3x decrease, from ~28 to ~10

but I think the major thing is more likely to be a reduction in the outliers than in the average

quartata

12:32 AM

Not much between 4 and 5 though

@ArtOfCode could you include the standard deviation as well

for each flag level

That would let people do actual tests

thesecretmaster

@ArtOfCode Maybe I haven't sat in CHQ for long enough, but that doesn't really seem like a useful improvement

ArtOfCode

56 secs ago, by ArtOfCode

but I think the major thing is more likely to be a reduction in the outliers than in the average

quartata

Hence why you need the standard deviation

pretty sure SQL has a builtin

make sure it's sample and not population stdev tho

since this is since the experiment thing started right

SmokeDetector

[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, potentially bad keyword in body, potentially bad keyword in title: Are all the holocaust witnesses jews or nazis? by user29558 on history.SE

K.Dᴀᴠɪs

I think OP deleted it... But it could have been a legit question imo

quartata

12:42 AM

sd k

No it's not.

Also it was deleted by a mod

K.Dᴀᴠɪs

I am doubtful that it was - but people ask the strangest things.

quartata

this idiot comes in and either posts flat out defamatory stuff or strawpeople about once a month

K.Dᴀᴠɪs

Oh I haven't noticed :)

quartata

it's why we have it blacklisted

on the bright side they were a little too obvious this time around and got autoflagged lol

we usually don't get two reasons on them

J F

1:01 AM

@SmokeDetector I’m pretty sure Anne Frank was Jewish.

Undo

1:14 AM

Sanity check: A camera's field of view increases linearly with distance from the target, right?

I.e. if I can see 5 feet horizontally at x feet away from the target, I should expect to see 10 ft at 2x away?

It makes intuitive sense (you can't bend light), and I have data in a scatter plot with r^2 = 0.9999, so... I would think so. But trig is scary.

@ArtOfCode Want to use that box plot I've been throwing in here occasionally?

thesecretmaster

You can bend light. That's what lenses do.

Undo

@thesecretmaster ... so I can't assume that and make math easy

thesecretmaster

I don't think it's linear

Undo

I mean you can't bend light outside of the lens. See what I'm saying? If it was anything other than linear, you could see around corners.

John Dvorak

/me declares Secret insane

thesecretmaster

1:22 AM

Took you that long?

I thought it was obvious from the start :P

Undo

22

A: How do I calculate the ground footprint of an aerial camera?

I got a little carried away with formatting my answer... This drawing is adaptable and can automatically calculate different scenarios, I'll give LaTeX/Python source to anyone who wants it. Edit: I've put the source code here. I must warn potential viewers that it's difficult to read and ...

Hah!

There we go.

SmokeDetector

[ SmokeDetector | MS ] Link at end of answer: Web browsers with NPAPI plugins support by CitoplasmaX on askubuntu.com

thesecretmaster

1:43 AM

@Undo Who can push to master for MS?

SmokeDetector

[ SmokeDetector | MS ] Manually reported answer: Do the trapped souls in the Reaper/Shinigami suffer? by Rich Chigga on anime.SE

tpu- by K.Dᴀᴠɪs

Undo

@thesecretmaster You probably need to be added to that list.

But it's me, Art, rest of Charcoal owners.

thesecretmaster

Charcoal owners == code admins?

Undo

Normal admins

Permissions are complicated :P

thesecretmaster

Was you+art, is now you+art+angus+triplee+one other person?

Undo

1:50 AM

Andy, yep

thesecretmaster

There was a list somewhere of people with privs? charcoal-se.org? I can't seem to find it now though

Undo

So yeah, metasmoke team is me and Art; tripleee, Andy, and angus inherit write privs from the org ownership

@thesecretmaster charcoal-se.org/people

That page needs some updating too

thesecretmaster

Could it be grabbed from MS with some JS?

Undo

We do that in some other places, yep

The site mod list

thesecretmaster

Also, why is MS not metasmoke.charcoal-se.org?

If someone here owns that domain?

Oh it is

Nevermind

Follow up: Why doesn't metasmoke.erwaysofsoftware 301 redirect to metasmoke.charcoal-se.org?

SmokeDetector

2:15 AM

[ SmokeDetector | MS ] Offensive title detected: How do I fuck the Instapussy? by Ash Ton on interpersonal.SE

quartata

IPS gets some weird stuff

SmokeDetector

[ SmokeDetector | MS ] Offensive answer detected: Is 'compromisation' actually used, or is there a better word? by luke on ell.SE

tpu- by K.Dᴀᴠɪs

Andy

2:33 AM

@ArtOfCode I am happy to post again

SmokeDetector

[ SmokeDetector | MS ] Username similar to website in answer: What is the best way to seal caulk tube? by tubeanew on diy.SE

[ SmokeDetector | MS ] Link at end of answer: Hiding secret from command line parameter on Unix by sambuca on stackoverflow.com

ArtOfCode

PostCount	FlagCount	ATTD		StdDev	CommonMax
55		1		190.5091	197.59	585.69
55		2		85.0182		109.86	304.74
62		3		48.9355		83.57	216.07
68		4		26.9559		51.98	130.92
56		5		10.2143		5.60	21.41

where CommonMax = ATTD + 2*StdDev

SmokeDetector

[ SmokeDetector | MS ] Bad NS for domain in body, bad keyword in body, bad keyword in title, blacklisted website in body: e most effective method to Remedy Acne by Smf odj on apple.SE

tpu- by K.Dᴀᴠɪs

ArtOfCode

i.e. 97% interval max

SmokeDetector

fp- by Makyen on Hiding secret from command line parameter on Unix [MS]

ArtOfCode

2:49 AM

@Undo wanna query prod for those stats? ATTD on 3 flags is 48.9, which seems like a pretty big jump

Undo

3:09 AM

@thesecretmaster erwaysoftware.com is my primary email domain; it's highly unlikely to ever go away. The charcoal domain doesn't have that guarantee, so we wouldn't want a bunch of links pointing to it.

@ArtOfCode What are you looking for?

ArtOfCode

26 mins ago, by ArtOfCode

PostCount	FlagCount	ATTD		StdDev	CommonMax
55		1		190.5091	197.59	585.69
55		2		85.0182		109.86	304.74
62		3		48.9355		83.57	216.07
68		4		26.9559		51.98	130.92
56		5		10.2143		5.60	21.41

experiment stats

just want to check that my query hasn't broken somehow

Undo

Why not just show the box plots?

(but yeah, can do)

Got the query?

ArtOfCode

SELECT
	COUNT(DISTINCT q.id) AS 'PostCount',
	MAX(q.flag_count) AS 'FlagCount',
	AVG(TIMESTAMPDIFF(SECOND, q.created_at, q.deleted_at)) AS 'ATTD',
	ROUND(STDDEV(TIMESTAMPDIFF(SECOND, q.created_at, q.deleted_at)), 2) AS 'StdDev',
	ROUND(AVG(TIMESTAMPDIFF(SECOND, q.created_at, q.deleted_at)) + (2 * STDDEV(TIMESTAMPDIFF(SECOND, q.created_at, q.deleted_at))), 2) AS 'CommonMax'
FROM (
	SELECT
		t.id,
		t.created_at,
		t.deleted_at,
		t.reason_weight,
		sites.max_flags_per_post,
		COUNT(DISTINCT flag_logs.id) AS 'flag_count'

Undo

55	1	190.5091	197.59	585.69
55	2	85.0182	109.86	304.74
62	3	48.9355	83.57	216.07
69	4	28.1304	52.50	133.14
56	5	10.2143	5.60	21.41

quartata

update_json works

now I have to wait for it to reload the subscriptions

watch the fetcher work like first try after spending 3 hours debugging the simple part

..well it started the timer

now we just need someone on IPS to post a comment

which might take a while since everyone's watching the oscars

SmokeDetector

3:15 AM

[ SmokeDetector | MS ] Repeating words in answer: What weed covering should I use under my mulch? by wow on gardening.SE

iBug

@Undo Are we going to contact a CM to fetch the actual TTL for each test post for us?

@SmokeDetector k

quartata

we already did that

Undo

@iBug Turns out our TTL data is very accurate for TPs. Shog's data verified that.

iBug

Or are we just going to use our (MS) data?

quartata

nothing has exploded yet? how is that possible?

Undo

3:16 AM

I'm comfortable trusting what we have. It paints a very believable picture with no overly weird outliers.

iBug

@Undo Good news.

Undo

@quartata It's saving up for a big explosion :P

iBug

But, does our data contain those systematic delays like the 3s wait for SEAPI?

Undo

It's measured from when we catch the post.

iBug

oh fine.

Undo

3:19 AM

You raise a decent point, probably good to investigate or throw an asterisk in there

Hey @Andy - what non-Excel thing do you use that I should totally be using too?

For graphs and such

Andy

matplotlib

python

Undo

Current data

@Andy Awesome, will look at for the future. Thanks.

ArtOfCode

@Undo possible to extend the whiskers to the 97% interval?

iBug

wolfram alpha

hahahaha

Undo

Maaaaybe

ArtOfCode

3:22 AM

and maybe international-ify the dates? :P

Undo

freedom dates

iBug

just use %Y-%m-%d %H:%M:%SZ

Undo

egads

iBug

like when you hover over a time on SO

quartata

...odd

so it's sending out the websocket hello

Andy

3:24 AM

Can we get it in stardates instead? That'd make the SFF people happy, I imagine.

iBug

or drop into SOCVR and check

quartata

sending data GET / HTTP/1.1
Host: qa.sockets.stackexchange.com
Upgrade: websocket
Connection: Upgrade
Sec-WebSocket-Key: ...
Sec-WebSocket-Version: 13

and then I don't get anything back

and because of that it never starts any of the pollers either, it waits for the socket to open

when it was running the first time it worked because it was checking Redis to reload the subscriptions (I had it already running when I updated the config)

but here it's not loading the subscriptions at the beginning

Undo

This is your individual websocket subscription project?

quartata

yes

it's almost working

ohh

I bet EventMachine doesn't know to do SSL

I put it on 443 and then forgot lol

SmokeDetector

[ SmokeDetector | MS ] Blacklisted website in answer: Windows Server 2016 Essentials and Virtualization rights by Lourdes Paul on serverfault.com

tpu- by K.Dᴀᴠɪs

tripleee

3:29 AM

!!/blacklist-keyword le\W?derme\W?luxe

SmokeDetector

@tripleee Blacklisted le\W?derme\W?luxe

quartata

yep there we go, needed a start_tls

thesecretmaster

@quartata I have a gem for the realtime socket, if you're interested

quartata

ok I swear remembering that the class is named Model and not Models will be the death of me

SmokeDetector

CI on d61fb5e succeeded. Message contains 'autopull', pulling...

[ SmokeDetector ] SmokeDetector started at rev d61fb5e (SmokeDetector: Auto blacklist of le\W?derme\W?luxe by tripleee --autopull) (running on teward/Solar Flare)

Restart: API quota is 18911.

quartata

3:33 AM

now I just have to wait 28.8 seconds

enthusing

> {"error_id"=>400, "error_message"=>"site is required", "error_name"=>"bad_parameter"}

:thonk:

I swore I put it in

hey on the bright side

ArtOfCode

Okay, box plot is in the draft... Anyone see anything else left to do before we post?

quartata

it pushed something onto a queue

eventmachine popped it off and called a callback

and it actually made an http request asynchronously

that's better than expected

SmokeDetector

tpu- by tripleee on How do I fuck the Instapussy? [MS]

fp by tripleee on Web browsers with NPAPI plugins support [MS]

tpu by tripleee on How do I fuck the Instapussy? [MS]

Undo

@ArtOfCode Does it mention that we're looking at the outliers more than the averages?

That's the best I can think up.

ArtOfCode

yup

Undo

3:41 AM

I'll probably put a better plot in there when I figure out matplotlib, that can be edited in

Andy

> casting between 1 and 5 flags randomly on each post eligible for flagging.

I thought it was only high weight posts

quartata

I'm using SNO correctly right: !parsed['items']&.empty?

it looks weird

Undo

300+, yes

ArtOfCode

@Andy good spot, fixed

Anything else? We ready to go?

quartata

oh woops

Undo

3:46 AM

I'll give it one more read, then yes

quartata

that's not what I want

it'll invert the nil

Andy

Give me just a bit to read over it, but I need to go do something first. I'll be back in 10-15 minutes.

Undo

Title could still use work. Ideas?

Andy

DIE, SPAM DIE!

Undo

First para: "We'd like to increase that to 4 or 5 automatic flags to reduce the time spam spends alive on the network." - why not say 5 and make it concrete?

We can always do 4 if we want to; don't want someone saying "meta approved 4 not 5"

quartata

3:48 AM

.present? and .blank? are so convenient

why are they only in Rails

Catija

You're going to get very little in the way of takers on five, I think... Just from listening to what people have said.

Undo

Four doesn't make sense in the same way 5 does.

I'll take on the doubters if I have that ridiculous pixel graph on my side.

^ Four looks awfully similar to 3. Five is a huge difference.

PeterJ

One thing about 5 is that it only needs a single person to fat-finger one and you won't have a chance to retract it.

Undo

we've been over this - it doesn't happen. You have to combine that 1 in 30k case with the chance of a human here making a mistake. I'll deal with that one time that happens in a decade.

SmokeDetector

[ SmokeDetector | MS ] Bad keyword with email in answer, email in answer: How do you get Kyogre in Omega Ruby? by Phoenix on gaming.SE

Catija

3:52 AM

It's not the difference between 10 minutes and 2... or something big.. it's between 2 and 1... Considering the post delay, with a question, most people won't see it either way.

quartata

look at the standard deviation though

the spread is what matters

it becomes much more consistent at 5

that's a good graph for it actually

Undo

Pretty high number of outliers for only ~60 datapoints in each bucket.

Catija

@SmokeDetector That... is actually an answer?... I think.

Undo

ah, Gaming

quartata

@SmokeDetector f

SmokeDetector

3:55 AM

fp feedback on autoflagged post: How do you get Kyogre in Omega Ruby? MS (@iBug @Floern @doppelgreener)

quartata

it's correct actually

just needs the email edited out

Undo

The capitalization is horrid, but yeah

Catija

... And all of the awful caps.

quartata

and maybe get the people with 158 reason weight conditions to re-evaluate their life choices

the new pings help a bit

but kinda getting old with the "two heavily correlated reasons FP" shtick

Undo

Those conditions are at least 99.5% accurate. If we want to change that, reevaluate that number

(which may be valid)

quartata

3:57 AM

it's not about the overall accuracy

Nisse Engström

@Catija "... I Got A Service To Get Any Pokemon..." Sounds suspicious.

quartata

@NisseEngström "as long as it ain't mythical" -- sounds like he's just using a gameshark and pokemon bank

(pokemon bank usually won't let you trade non-event ones)

Catija

I suggested an edit.

K.Dᴀᴠɪs

Why not base the number of autoflags to the accuracy... 5 = 99.75%, 4 = 99.50%.. I feel there will be less debate on up-ing the flag count if we can say we are even more sure the post is spam.

Undo

Oh, that's what we're proposing.

Transcript for