« first day (1562 days earlier)      last day (2244 days later) » 

12:01 AM
204 out of 51564 don't have a deletion date in the dump for one reason or another (some of those correct). Not going to worry about 0.3%
-10	385
-9	408
-8	453
-7	473
-6	523
-5	658
-4	744
-3	909
-2	1157
-1	2946
0	22865
1	7325
2	461
3	203
4	155
5	103
6	101
7	114
8	71
9	83
10	77
Frequency of deltas between -10 and 10 (seconds, actual - ours)
Full version of the above table; 0 is at L957 this is borked
So we have very accurate numbers for more than half of the 51360 posts where the following is true: (1) MS had deletion information, and (2) Shog's dump had deletion information
Now I'm going to look at one of those +/- 10 second delta posts and see what's going on. Hoping it's clock skew that can be corrected historically without changing our systems.
MS post 46907 is stackoverflow.com/a/40637555. MS records deletion at 2016-11-16 16:53:10, Shog's dump claims deletion at 2016-11-16 16:53:20 (10 seconds later). Shog is right according to the web interface.
Nothing unusual in the post timeline
Report is this:
Nov 16 '16 at 16:52, by SmokeDetector
[ SmokeDetector | MS ] Manually reported answer: How to improve performance of map that loads new overlay images by Francisco on stackoverflow.com (@dorukayhan)
Hah
Chat report was at 16:52:55Z. MS creation date is 16:52:47
darn near a 10 second delta
Real creation date on site was 16:41:39 - we were slow on that one, but it doesn't really matter.
 
SE was having garbage websocket issues that day
 
k, we'll look at a later one
going out to 9-11 second delta to get a recent example
well that's interesting
There are only 3 in that range since 2016-11-16 19:31:06 (MS 46919)
All on sites I don't have 10k for
 
What are the sites?
 
Apple, Cooking, Money
MS 76284, 84703, 101895
 
@Glorfindel
 
12:16 AM
I don't really need 10k, I suppose. I trust Shog's dump.
Jul 14 '17 at 8:24, by SmokeDetector
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, blacklisted website in body, pattern-matching product name in body, pattern-matching website in body, +1 more: It allows to beautify the dimensions of the penis and makes it more strong by eamber juri on apple.SE (@CalvT븃)
8:24:25Z actual report; 08:24:21Z MS creation date
Within tolerances on that end
Looking at all deltas between 5 and 30 seconds. There are only a few after 47xxx series IDs.
26
1107 / 1136 occurred before 2016-11-28
I'm going to limit my analysis on these to posts after 01/01/2017
oh what the heck
apparently subtracting dates in MySQL isn't that simple
2018-02-15 13:12:59 - 2018-02-15 13:13:00 = -41
not buying it
Okay. With that fixed....
Running the previous report, but with fixed timestamps and only posts after 2017-01-01
-10	2
-9	1
-8	1
-7	1
-6	2
-5	1
-4	9
-3	117
-2	238
-1	6430
0	17734
1	1784
2	994
3	820
4	667
5	525
6	401
7	431
8	400
9	368
10	377
An interesting - and encouraging - sudden dropoff in the negatives
That pattern continues - a long tail of singles in the negatives, while the positives have a long tail of not-singles
The timestamp fix brings up way more posts where we're late with deletion logs
 
inevitable
 
101661 / stackoverflow.com/questions/48229219 - we're ten seconds late with the deletion log
Jan 12 at 15:25, by SmokeDetector
[ SmokeDetector | MS ] Offensive body detected: How can you get current packages for linux? by user2762996 on stackoverflow.com (@iBug @TetsuyaYamamoto)
But on that one, the MS report is created waaaaay before the post here
15:25:15 MS, 15:25:56 chat
MS clock is currently in sync, as far as I can tell
 
12:32 AM
[ SmokeDetector | MS ] Repeating words in answer: On an MSI laptop, how do I disable the backlite keyboard by user799607 on askubuntu.com
 
sd n
@Undo Could just be chat ratelimiting on that one.
MS 106659 / stackoverflow.com/a/48806358 was deleted twice - we were right on the first deletion; Shog's dump takes the last one.
 
Art's Midnight Musings #2: Apparently I am writing that paper now... not quite sure how that happened, but apparently the fingers decided to type and here we are
 
@Undo mod link
 
105670 / stackoverflow.com/a/48681086 was off by 12s. Looks like a natural delay
These are insanely rare recently. I think our deletion logs are actually very solid.
Outside of the ones we're just missing, of course. quartata's fixes should make that way better.
Still, though, I think we have very good log data on this. Rows and rows and rows of 0/1/-1 deltas with Shog's data.
 
12:50 AM
@Undo sounds good... next steps/improvements, etc?
 
@ArtOfCode I'm thinking the next step is running a rand(1, 5) flag count experiment on posts with score >= 300. See any problems with that?
Run it for a week or so, collect data.
 
You wanna put the meta post out before or after that?
 
after
 
Okay
No problems per se, but probably a couple things to think about
 
If we do it before, we're going to be fighting the "but you have no data" crowd. I think I'd rather fight the "site agency" crowd.
 
12:52 AM
whatcha wanna do about core flag count?
 
@ArtOfCode I was thinking we use core for the first 2 flags, then fill with commoners past that
 
heh
yeah, works for me
 
Was going to add a FlagSetting for five_flag_experiment_weight - set to ten thousand to turn it off
 
When we do the actual 5-flags we should probably scale core flags appropriately
 
Definitely
 
12:54 AM
@angussidney yeah, just flag_count / 2
 
rounded up like now
 
aye
 
so 3/2 for those
 
Okay. Probably a good thing I put better log tooling in before we do this :P
 
eh, shouldn't be dangerous. We'll review it before I fat-finger new SO questions into oblivion.
(not saying that'd be a bad thing...)
 
12:56 AM
@Undo you want me to leave writing the code to you for this or do I just go ahead?
 
I'll take a stab at it, unless you really want to
 
Nah, go for it. Just working out whether to get going now or let you do it 'cuz you know how you want it to work ;)
 
I got it ;)
you have a paper to write :P
 
my hobby is giving myself new things to do and never finishing them :P
6
I also have haskell to learn, Java to write, and JS to finish, but I'm messing around with papers instead :)
 
Haskell? School thing?
 
1:03 AM
Aye
may have missed the fact that our coursework is in haskell for the past three months and now have three weeks to learn it and do coursework
 
hah
 
֑
 
I don't get why people use shortened URLs for legit sites here.
 
1:27 AM
[ metasmoke ] ci/circleci failure on b1da0b6: Your tests failed on CircleCI
[ metasmoke ] ci/circleci failure on f877204: Your tests failed on CircleCI
 
@ArtOfCode let me know if you have questions
 
What is life?
How big is the universe?
What is the largest object known to man?
oh haskell questions :)
probably many
 
[ metasmoke ] ci/circleci failure on 5be5519: Your tests failed on CircleCI
 
@Undo MS record 108024 was created for testing purposes. Good to destroy?
 
You sure?
 
1:37 AM
looks like a TP to me
 
Yes. I was testing my new command.
 
o.O
 
See the last line of "why".
 
still caught a useful post, though :P
 
oh, I see. Duplicate
 
1:38 AM
don't we cleanup tests?
 
sometimes
 
yeah. @ArtOfCode he ran it on an already-caught post.
108022 was the original
 
@Undo aye, I hit the domain list
 
@Undo If you search the context you'll find I !!/scanned two new reports
 
don't destroy
 
1:38 AM
gone
 
you didn't need to do that fwiw
 
... too late
 
I had already tested it on real posts
 
Strangely, one got "does not look like spam", the other went to 108024
 
actually, never mind :P. @ArtOfCode review results keep us from nuking posts.
 
1:39 AM
need a dependent
 
@Undo hehe. I got it from local cache, anyway.
 
@quartata Didn't know that, thanks anyway
 
Anyway, there might be a bug there somewhere
those two records have different weights
same text
 
blacklisted user
 
oh
rolls eyes
 
1:44 AM
[ metasmoke ] ci/circleci error on dfa468c: Your CircleCI tests were canceled
[ metasmoke ] ci/circleci error on dfa468c: Your CircleCI tests were canceled
[ metasmoke ] ci/circleci success on dfa468c: Your tests passed on CircleCI!
 
\o/
 
There we go. When you get a chance, github.com/Charcoal-SE/metasmoke/pull/331
 
Heh, the comparison was already open in another tab :P
 
8
Q: A clarification request for declined spam flag

A JI flagged this post for obvious spam link inserted in the end when posted in Charcoal HQ. Check the first version here. After looking at the user's profile and the link in the end, I flagged it as spam. I also posted a comment asking the user to state any affiliation to the site. In reply to tha...

Never tried it, but it looks promising
 
2:03 AM
[ SmokeDetector | MS ] Bad NS for domain in answer: Remove / Rename Uncategorized Category In Wordpress by Sahadath Hossain on wordpress.SE
[ metasmoke ] ci/circleci failure on b453c16: Your tests failed on CircleCI
 
>:(
 
Bahahaha one char over :D
 
[ metasmoke ] ci/circleci success on 4fbb960: Your tests passed on CircleCI!
 
Gonna deploy that tonight?
...says he, as it's already deployed
 
So it's disabled. Whoever is awake when we start getting spam next can try it out. Probably me or you
 
2:14 AM
[ metasmoke ] ci/circleci success on b15ce72: Your tests passed on CircleCI!
 
@Undo spam hour's only a couple hours out
 
I should be around
 
also I may have broken metasmoke... how big is the prod log at the moment?
 
Usually weekends are iffy
 
@Undo Monday morning should be okay
 
2:16 AM
1.7G
iffy for spam volume, I mean. Not my schedule.
 
@Undo yeah, maybe grepping through that wasn't the best idea I've ever had
Miiiight have to pipe tail output to it first
 
hah
tail -n 10000 or something
 
aye
 
In fairness, it shouldn't be that big. I just can't figure out why s3cmd doesn't want to work.
 
uh
bahahahaha
you're gonna need to kill the server :D
that request is literally never going to return because it's waiting on the command line for input
grep -E '/api/v2.0/posts/[0-9]+/flag' -C 30 --color=never
 
2:23 AM
Restart: API quota is 17126.
fp by ArtOfCode
 
Restarted nginx
 
tpu- by Tetsuya Yamamoto
 
@Undo Oughtta do it. There's a commit there to make that not happen now :)
I think that takes the cake for the most I've managed to brick metasmoke
 
At least grep hanging isn't really resource heavy :P
deploying
 
[ metasmoke ] ci/circleci success on 76f18a5: Your tests passed on CircleCI!
 
2:26 AM
done
 
@ArtOfCode just wondering (for my own personal use) how much were the stickers? (I'm wondering what specs you did as well)
 
@WELZ in total? Both orgs?
 
just one I guess
 
@WELZ one org would've been £90
 
how many stickers was it?
 
2:33 AM
340 per org
Total was £157.61, because they give you a discount for doing multiple designs in one order
 
A J
@K.Dᴀᴠɪs Thanks will check it for sure.
 
and they were 2x2 die cut stickers?
 
@WELZ 5x5 cm
so yeah, 2x2 in
 
@ArtOfCode Is that minimum order?
 
They adjusted the sizes slightly to get the proportions right, but at least one side is 50mm
 
2:35 AM
@Undo no, it was just extra for money left
 
@Undo nah, minimum is 10, 340 is just what we had the money for
 
minimum is 10
 
Does it scale linearly down to 10?
That'd surprise me
 
Cost? Nope. Dunno what the algorithm is, but it's not linear
 
Yeah. They have tooling cost or something
 
2:36 AM
£13 for 10
 
The more you order, the less you pay per sticker. (as with just about everything)
 
Probably have to make the die to cut them.
 
or $17 usd
 
(but if you want 10 just get a sample pack for about £7)
 
How many sample packs can you get? :P
 
2:37 AM
50 is £46, 100 is £55
@Undo Not more than one for the same design in one order, at least :P
 
[ SmokeDetector | MS ] Bad keyword in title: how to Watch The Walking Dead Season 8 Episode 8 Online by nafodis on scifi.SE (@Mithrandir)
tpu- by WELZ
 
aha! this wasn't caught for sites.google...
 
2:57 AM
 
fp- by Tetsuya Yamamoto
 
@SmokeDetector f
 
@iBug it was just funny that right after being removed... (at least it was caught for something else - though it was obviously spam, so prob would've been caught anyhow)
 
The Chinese name of Winnie the Pooh (a Disney cartoon character, I bet you all know) is banned in China (blacklisted keyword)
 
3:24 AM
@SmokeDetector my main partition goes skraaaaaaa
 
@iBug Just banned in mainland or includes HK too? Do you know which reasons behind banning that term?
 
Rob
3:35 AM
@iBug Hmm.. and to think, I get frustrated at my ISP for throttling youtube at peak time :\
Or more dishonestly.. unthrottling all data to speed tests
@iBug A couple options. One could register their FKey with MS, rather than just the oauth token. Or potentially less dangerous; a userscript which detects flagged posts which need to be retracted
Granted, the second option requires people to be online
 
@Rob Just a machine which is on and has an internet connection through which they are logged in, which could be a mobile device. Doing it on the user's equipment isn't as convenient as having it done via MS, but it's certainly not that difficult.
 
[ SmokeDetector | MS ] Bad keyword in title: How to host machine learning in alibaba cloud by Bhargava Ganti on stackoverflow.com
fp- by K.Dᴀᴠɪs
 
Rob
@Makyen Well, yeah, that could work. Though then we'd need mobile apps, desktop apps, and so on. Rather than just a userscript
And a further alternative, if we're worried about nuking non-spam, would be to post a message to a mod chat room, so that the flags can be declined/cleared
 
3:52 AM
[ SmokeDetector | MS ] Bad NS for domain in body, bad keyword in body, blacklisted website in body, potentially bad keyword in body: Passcode lock feature to prevent iPhone from being by Shf ldvhj on apple.SE
 
@Rob fkey also requires cookies. It's not feasible
 
@Rob You can use userscripts with Firefox for Android.
 
tpu- by Makyen
 
Rob
@Undo Ah, right
 
[ SmokeDetector | MS ] URL in title, bad keyword in body, bad keyword in title, blacklisted website in body, blacklisted website in title, +1 more: fornatgaex.com/primal-alpha-beast/ by spmlxhgvap on meta.SE
tpu- by Makyen
 
4:03 AM
Turned on five flag experiment. If it goes nuts, !!/stopflagging
13
Should flag posts over weight 300 with a random number of flags between 1 and 5, in an attempt to collect deletion time statistics.
 
[ SmokeDetector | MS ] Mostly non-Latin answer: ASP.NET MVC 3 Adding comments in article view by Ruchi Kumari on stackoverflow.com
tpu- by Rob
[ SmokeDetector | MS ] URL-only title, bad keyword in body, bad keyword in title, blacklisted website in body, blacklisted website in title, +3 more: realcoloncleansingworks.com/andras-fiber/ by guro on askubuntu.com
tpu- by Undo
fp- by Makyen
[ SmokeDetector | MS ] URL-only title, link at end of body: fitnessexperttips.com/elite-biotics/ by warunqwik on askubuntu.com
tpu- by Makyen
[ SmokeDetector | MS ] Bad NS for domain in body, bad keyword in body, blacklisted website in body, pattern-matching website in body, potentially bad keyword in body: ny way to Add a reminder more t ahead of a Goo event by user179510 on webapps.SE
tpu- by Makyen
 
4:26 AM
@SmokeDetector Four flags on that one
11 second deletion time
 
@Undo Yep. It was gone well before I provided feedback/tried to flag.
 
@Undo Testing or random mistake?
 
Testing
autoflagging doesn't make random mistakes :)
 
@Undo Are posts randomly chosen for testing, or scheduled?
 
weight >= 300; rand(1..5) flags
 
4:31 AM
All that are over 300 are chosen?
 
yes
 
Just curious. Is there any particular benefit of using a random number of flags?
 
@K.Dᴀᴠɪs To collect the data of the correlation between flag count and TTL.
 
Want to get numbers on deletion time relation to flag count. We don't have statistically valid data on that across large numbers of posts.
Best we have is from the 0 -> 1 and 1 -> 3 moves, both time-delimited.
 
That makes sense
Are these flags manually deleted?
 
4:35 AM
Hm?
 
@TetsuyaYamamoto Probably HK not affected. It's because Winnie is "visually similar" to the current chairman, whose proposal to remove the limit of consecutive terms that a chairman can serve for.
People have used Winnie to refer to the chairman, Xi, as a circumvention of keyword filters.
 
[ SmokeDetector | MS ] URL-only title, bad keyword in body, bad keyword in title, blacklisted website in body, blacklisted website in title, +4 more: nutritionofhealth.co.uk/total-age-repair/ by user799642 on askubuntu.com
tpu- by K.Dᴀᴠɪs
 
@SmokeDetector One flag, 23 seconds
 
[ SmokeDetector | MS ] Bad NS for domain in body, bad keyword in body, bad keyword in title, blacklisted website in body, link at end of body, +2 more: Your body renews every day a kilogram of muscle by ubxkgcqcu on askubuntu.com
tpu- by K.Dᴀᴠɪs
 
4:59 AM
@Undo How are flagging candidates chosen?
We have 2 core + 1 any for 3, but what about 1 2 4 5?
 
[ SmokeDetector | MS ] Bad NS for domain in body, bad keyword in body, bad keyword in title, blacklisted website in body, link at end of body, +1 more: Enduraflex Canada Colored contact by andrflt12 on askubuntu.com
[ SmokeDetector | MS ] URL in title, pattern-matching website in body, pattern-matching website in title: wellnessdietsolution.com/elite-biotics/ by Rinengex on english.SE
tpu- by Makyen
 
ELU and Workplace also in test?
 
every site.
oh, well, not those.
 
5:06 AM
yeah
 
[ SmokeDetector | MS ] Blacklisted username, offensive answer detected: Potentiometer to change voltage? by niggery on electronics.SE
 
!!/watch 9785347272
 
@tripleee Added 9785347272 to watchlist
[ SmokeDetector | MS ] URL in title, link at end of body: fitnessexperttips.com/elite-biotics/ by Tayastillo on drupal.SE
tpu- by tripleee on Potentiometer to change voltage? [MS]
tpu- by tripleee
CI on c26d06d succeeded. Message contains 'autopull', pulling...
[ SmokeDetector ] SmokeDetector started at rev c26d06d (SmokeDetector: Auto watch of 9785347272 by tripleee --autopull) (running on teward/Lunar Eclipse)
Restart: API quota is 16187.
 
!!/watch champcash
 
@tripleee Added champcash to watchlist
[ SmokeDetector | MS ] Bad NS for domain in body, bad keyword in body, blacklisted website in body, pattern-matching website in body, repeated URL at end of long post: Spread your protein over 3 to 4 shots a day and you're done! by ubxkgcqcu on apple.SE
tpu- by K.Dᴀᴠɪs
 
5:18 AM
!!/watch fitnessexperttips\.com
 
@tripleee Added fitnessexperttips\.com to watchlist
 
!!/blacklist-keyword elite\W?biotics
 
@tripleee Blacklisted elite\W?biotics
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, blacklisted website in body, link following arrow in body, pattern-matching website in body: Tryvix Creme, at that point you will be happy by Arleneberg on graphicdesign.SE (@TetsuyaYamamoto)
tpu- by K.Dᴀᴠɪs
CI on d135020 succeeded. Message contains 'autopull', pulling...
Restart: API quota is 16169.
 
So in the event an autoflagger is not set up for a site, how does that work?
Does MS throw an exception? Does it think it went through?
 
The API throws an error and MS just tries to flag using a different account
 
5:26 AM
@K.Dᴀᴠɪs Use another flagger
 
5:37 AM
[ SmokeDetector | MS ] Bad keyword with a link in answer, link at end of answer: What do you call these snips/shears? by Mike on crafts.SE
tpu- by tripleee
 
@SmokeDetector k
 
[open] Hi o/
 
!!/watch forgica\.com
 
@tripleee Added forgica\.com to watchlist
CI on 23e4286 succeeded. Message contains 'autopull', pulling...
[ SmokeDetector ] SmokeDetector started at rev 23e4286 (SmokeDetector: Auto watch of forgica\.com by tripleee --autopull) (running on teward/Lunar Eclipse)
Restart: API quota is 15962.
 
5:54 AM
how do you pronounce git
"dgit" or "ghit"?
 
[ SmokeDetector | MS ] Pattern-matching website in body: Getting scraped data using api with python by Ismail Chouiti on stackoverflow.com
fp- by tripleee
 
@iBug your orthography doesn't look correct for how I pronounce it. In Italian /gh/ is used to mark a hard g (the normal is for g to be soft before e and i) but in some other languages it specifically spells a soft g, or an uvular R, or a number of other sounds ... but I guess this is what you are getting at?
hard g, like the English word git
merriam-webster.com/dictionary/git has a speaker icon you can click on
 

« first day (1562 days earlier)      last day (2244 days later) »