« first day (2504 days earlier)      last day (1306 days later) » 

9:05 AM
[ SmokeDetector | MS ] Bad ip for hostname in body, bad keyword in body, bad keyword in title, bad pattern in url body, pattern-matching product name in body, +3 more (484): How Does Keto Premiere in Australia Work? by ketoprmre on meta.SE
 
tpu- feedback received
 
[ SmokeDetector | MS ] Blacklisted user (71): pycharm console errror, console not starting properly by Muzo on stackoverflow.com
 
9:27 AM
[ SmokeDetector | MS ] Bad ip for hostname in answer, phone number detected in answer, potentially bad ns for domain in answer, potentially bad keyword in answer, potentially problematic ns configuration in answer (191): Which Stack Exchange sites weren't proposed in Area 51? by Hulu Activate on meta.SE
 
tpu- feedback received
 
[ SmokeDetector | MS ] Bad pattern in url answer (63): Delete git.io shortened URL by yamiacat on stackoverflow.com
[ SmokeDetector | MS ] Bad keyword in link text in answer (81): Celery worker displaying "unknown option -A" on windows by Kev on stackoverflow.com
 
fp- feedback received
 
[ SmokeDetector | MS ] Potentially bad ns for domain in answer, username similar to website in answer (64): Angular sce.trustAsHtml not working by Games4html5 on stackoverflow.com
 
!!/blacklist-website wwwhulucomactivate\.com
 
tpu- feedback received
 
[ SmokeDetector | MS ] Bad keyword in body, link at end of body, pattern-matching website in body (290): what is mean of service center? by shekar pallegoni on superuser.com
 
tpu- feedback received
 
!!/watch lgservicecenterbangalore\.co\.in
 
@VictorVosMottor That pattern looks like it's already caught by Bad keyword in body and Bad keyword in answer; append -force if you really want to do that.
 
9:45 AM
!!/help
 
@VictorVosMottor I'm SmokeDetector, a bot that detects spam and offensive posts on the network and posts alerts to chat. A command list is available here.
 
tpu- feedback received
 
!!/watch jok3rshop
 
@VictorVosMottor You don't have blacklist manager privileges, but I've created PR#4927 for you.
 
PR#4927 ("Victor VosMottor: Watch jok3rshop") opened by SmokeDetector
 
9:50 AM
!!/watch 097653
 
@VictorVosMottor You don't have blacklist manager privileges, but I've created PR#4928 for you.
 
PR#4928 ("Victor VosMottor: Watch 097653") opened by SmokeDetector
 
!!/watch jok3r\.us
 
@VictorVosMottor The regex contains an unescaped "."; in most cases, it should be "\."; append -force if you really want to do that.
@VictorVosMottor You don't have blacklist manager privileges, but I've created PR#4929 for you.
 
PR#4929 ("Victor VosMottor: Watch jok3r\.us") opened by SmokeDetector
 
9:51 AM
[ SmokeDetector | MS ] Potentially bad ns for domain in answer (1): How do I implement async generators? by chuan zhang on stackoverflow.com
 
fp- feedback received
 
[ SmokeDetector | MS ] Potentially problematic ns configuration in answer (1): Fastest way to find modular multiplicative inverse by user828848 on math.SE
[ SmokeDetector | MS ] Potentially bad ns for domain in body, potentially bad keyword in body (2): How do i manage my expenses? by BUKU on meta.SE
 
tpu- feedback received
 
[ SmokeDetector | MS ] Mostly punctuation marks in answer (36): Use Python to launch Excel file by Finance_Professional on stackoverflow.com
[ SmokeDetector | MS ] Potentially bad keyword in answer (1): How to fix "Package is in a very bad inconsistent state" error? ✏️ by darkByt3 on askubuntu.com
[ SmokeDetector | MS ] Potentially problematic ns configuration in body (1): Getting a Compile Error: Sub or function not defined error in VB6 ✏️ by RishabhSh on stackoverflow.com
 
10:20 AM
[ SmokeDetector | MS ] Few unique characters in answer, luncheon meat detected, no whitespace in answer, potentially bad keyword in answer, repeating characters in answer (306): #JS (Mouse tracker) Relative coordinates depending on screen size by ILKetchup on stackoverflow.com
 
10:37 AM
[ SmokeDetector | MS ] Potentially bad keyword in answer (1): Does backing up WhatsApp on Google Drive expose messages? by Sumanth V S on security.SE
[ SmokeDetector | MS ] Url-only title, bad ip for hostname in title, bad keyword in body, bad keyword in title, blacklisted website in title, +3 more (594): slimketoboost.com/keto-premiere-dischem-za/ by mhukgkophmgu on askubuntu.com
[ SmokeDetector | MS ] Potentially bad ns for domain in body (1): Zabbix : Trigger default check time by Veljko Petrović on serverfault.com
[ SmokeDetector | MS ] Bad ip for hostname in body, bad keyword in body, bad keyword in title, blacklisted website in body, body starts with title and ends in url, +5 more (690): Keto Premiere Dischem South Africa by parrenwarrear on askubuntu.com
 
10:53 AM
tpu- feedback received
 
[ SmokeDetector | MS ] Potentially bad keyword in answer, username similar to website in answer (64): Cannot edit SiteMap in Dynamics CRM 365 by Talha Manzoor on stackoverflow.com
 
naa- feedback received
 
[ SmokeDetector | MS ] Bad keyword with a link in answer, potentially bad ns for domain in answer (82): Server Error in '/KenticoCMS' Application. Clicking on Tools in CMS Desk in Kentico by growthly zet on stackoverflow.com
 
11:09 AM
tpu- feedback received
 
11:23 AM
[ SmokeDetector | MS ] Potentially bad keyword in answer (1): Discord.js Music Bot ffmpeg not found? by Thiago Imai on stackoverflow.com
 
11:43 AM
There's been no scan activity for 3 status pings. There may be a problem.
There's been no scan activity for 4 status pings. Going to reboot.
[ SmokeDetector ] SmokeDetector started at rev 6995a52 (SmokeDetector: Auto blacklist of wwwhulucomactivate\.com by tripleee) (running on teward/Osiris, Python 3.6.9)
Restart: API quota is 14361.
[ SmokeDetector | MS ] Few unique characters in body, no whitespace in body, repeating characters in body (262): please help . i want it to be, if its not a prime number, it alerts n if it is, it alerts.using break statement in js thank you by Valentina Oni on stackoverflow.com
 
tpu- feedback received
 
[ SmokeDetector | MS ] Potentially bad ns for domain in body, potentially bad keyword in body (2): How can I pass the VMware 2V0-21.20 exam easily with questions? by kalvinluise894 on superuser.com
 
11:58 AM
tp feedback received on [MS] Cannot edit SiteMap in Dynamics CRM 365
 
@VictorVosMottor I "upgraded" my feedback from link-only NAA to spam based on the similar username to the website in metasmoke.erwaysoftware.com/post/266412
!!/watch quellxcode\.com\.au
 
@JeffSchaller You don't have blacklist manager privileges, but I've created PR#4930 for you.
 
PR#4930 ("Jeff Schaller: Watch quellxcode\.com\.au") opened by SmokeDetector
tp feedback received
 
sdc approve 4930
 
12:35 PM
[ SmokeDetector | MS ] Link at beginning of answer (33): How do I find the nearest prime number? by Pedro Hugo García Peláez on stackoverflow.com
 
@SmokeDetector Undisclosed affiliation.
 
tpu- feedback received
 
fp- feedback received
 
Hi @Spevacus
 
12:54 PM
@VictorVosMottor Why is that on the keyword list, not on the number list?
 
[ SmokeDetector | MS ] Potentially bad keyword in body (1): Could AI help fight with Covid-19? by Jim Innovator on stackoverflow.com
 
tp feedback received
fp feedback received
 
@VictorVosMottor Hello!
 
[ SmokeDetector | MS ] Bad keyword with email in body, potentially bad keyword in body, potentially bad keyword in title, potentially problematic ns configuration in body (60): Why lowest price upvc dealers in bangalore? by upvc door window suppliers on superuser.com
 
tpu- feedback received
 
12:58 PM
!!/watch- bangaloreupvcwindows\.in
 
PR#4931 ("Spevacus: Watch bangaloreupvcwindows\.in") opened by SmokeDetector
 
!!/watch- bangaloreupvcp(?:@gmail\.com)?
 
PR#4932 ("Spevacus: Watch bangaloreupvcp(?:@gmail\.com)?") opened by SmokeDetector
 
sdc approve 4931
 
1:01 PM
sdc approve 4932
 
1 message moved to Trashcan
 
[ SmokeDetector | MS ] Link at beginning of answer (33): Implement 3 finger drag by Matthew T Wetzel on askubuntu.com
[ SmokeDetector | MS ] Potentially bad keyword in body (1): How to install Pentablet Divers for XP Pen Deco 01 on Ubuntu 18.04? by user312195 on askubuntu.com
 
fp feedback received
 
[ SmokeDetector | MS ] Potentially bad ns for domain in answer (1): QML ComboBox item DropDownMenu style by ivand on stackoverflow.com
 
1:07 PM
fp feedback received
 
@Mast I thought number list is only for phone-numbers, ain't it?
 
fp- feedback received
 
!!/bisect FilmShowing { private Integer id; private Film film
 
@Spevacus Matched by (?:films?|complets?|regarder(?:s|es)?|t[ée]l[ée]chargements|en\s*ligne|gratuits?|diffusion|courant)(?:[\W_]*+\w+){0,5}(?:[\W_]*+(?:films?|complets?|regarder(?:s|es)?|t[ée]l[ée]chargements|en\s*ligne|gratuits?|diffusion|courant)){2,}(?#French verion) on line 15433 of watched_keywords.txt
 
1:14 PM
ಠ_ಠ
 
[ SmokeDetector | MS ] Link at end of answer, potentially problematic ns configuration in answer (62): How can I Edit WordPress Page in IDE like PhpStorm or Sublime? by Stas Ustimenko on wordpress.SE
 
naa- feedback received
fp feedback received
 
1:38 PM
@SmokeDetector Mais ouis
 
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, blacklisted website in body, potentially bad ip for hostname in body, potentially bad keyword in body (294): Okinawa Flat Belly Tonic Diet Easy Weight Losing by Okinawa Flat Belly Tonic natur on meta.SE
 
@Mast Mais oui---s---
 
tpu- feedback received
 
!!/watch- hbloom\.freshdesk\.com
 
1:41 PM
!!/watch maptia\.com
 
@VictorVosMottor That pattern looks like it's already caught by Blacklisted website in body and Blacklisted website in answer; append -force if you really want to do that.
[ SmokeDetector | MS ] Potentially bad ns for domain in answer (1): Who is Suvarchala? Is she wife or daughter of Sun God? by Priyansh Jain on hinduism.SE (@Pandya)
 
fp- feedback received
 
[ SmokeDetector | MS ] Potentially bad ip for hostname in answer, potentially bad keyword in answer (2): Why are glasses considered hats? by Mary_Smith2608 on meta.SE
 
tpu- feedback received
 
!!/watch- syghtglass\.com
 
1:48 PM
PR#4933 ("Spevacus: Watch syghtglass\.com") opened by SmokeDetector
 
!!/watch syghtglass
 
@VictorVosMottor You don't have blacklist manager privileges, but I've created PR#4934 for you.
 
PR#4934 ("Victor VosMottor: Watch syghtglass") opened by SmokeDetector
 
2:03 PM
!!/approve 4933
 
tpu- feedback received
 
!!/watch- viral481\.com
 
!!/bisect maptia.com
 
2:05 PM
@Mast Matched by maptia\.com on line 2869 of blacklisted_websites.txt
Makyen/EC2-linux: Executing automatic scheduled reboot.
[ SmokeDetector | MS ] Bad keyword in body, potentially bad keyword in body, potentially bad keyword in title (97): How To Use Swolgenix XL Light? by swolgenixbuy on patents.SE
 
tpu- feedback received
 
2:22 PM
[ SmokeDetector | MS ] Potentially bad keyword in body (1): Chartjs Stacked bar chart problem in total ✏️ by Majed Alomary on stackoverflow.com
 
fp- feedback received
 
I think I can modify the existing streetinsider blacklist to be better.
Not now.
 
fp- feedback received
 
2:36 PM
fp- feedback received
 
[ SmokeDetector | MS ] Mostly punctuation marks in title (27): 9560533808>\#+:#*9-#\_$.@ output by Gavin S on stackoverflow.com
 
fp- feedback received
 
@Xnero I'm sorry but it was made extremely clear to you that you were no longer welcome to participate in Charcoal last night and your continuing refusal to accept that and your continuing insistence on posting in this room after having been told multiple times that doing so is an abuse of your moderator abilities.
@Xnero If you want to continue this, send in an email to Support. I'm done.
 
[ SmokeDetector | MS ] Potentially bad ns for domain in answer (1): Bootstrap UI Angular modal doesn't show as a pop up by Dev Team on stackoverflow.com
 
tpu- feedback received
 
2:51 PM
[ SmokeDetector | MS ] Potentially bad asn for hostname in body (1): How do I add an "include" folder in my CMakeLists.txt? ✏️ by Jun_Hao on stackoverflow.com
 
@DanielWiddis What I've wanted to do since forever is explore how reasons become more/less accurate over time, and whether we can/should consider that in autoflagging weight
Spam patterns change, and the reasons themselves change. Seems like it's definitely be suboptimal to always consider the entire accuracy history of a reason
 
Might be as simple as "only look at the last n months or m posts" or maybe a more complex weighting system
 
3:03 PM
tp feedback received
fp feedback received
 
@metasmoke misclick
 
[ SmokeDetector | MS ] Potentially bad keyword in answer (1): Calendly Api integration by Vitzkrieg on stackoverflow.com
[ SmokeDetector | MS ] Link at beginning of answer (33): Tensorflow not getting installed with pip by Çağatay Şen on stackoverflow.com
 
fp- feedback received
 
[ SmokeDetector | MS ] Offensive answer detected, toxic answer detected (158): TDAmeritrade "Purchase Price" not matching fulfilled order purchase price by user102816 on money.SE
 
naa- feedback received
 
3:13 PM
[ SmokeDetector | MS ] Bad keyword in body (95): Galaxy s2 stuck in flight mode by abrackers on android.SE (@AndrewT.)
 
@SmokeDetector wut?
 
@SmokeDetector ignore-
 
3:34 PM
[ SmokeDetector | MS ] Potentially problematic ns configuration in body (1): PHP Cookies do Not work on Chrome But works on IE by pourya rajabi on stackoverflow.com
[ SmokeDetector | MS ] Potentially problematic ns configuration in body (1): Using az vm run-command from within a .ps1 script and with .ps1 script file by Petter Frykman on stackoverflow.com
 
fp feedback received on [MS] Galaxy s2 stuck in flight mode
fp feedback received
 
[ SmokeDetector | MS ] Link at end of answer (61): Can I monitor my phones temperature? by T Anandam on android.SE (@AndrewT.)
 
naa- feedback received
 
3:49 PM
They hit the go button a little too early.
 
tpu- feedback received
 
4:06 PM
@Spevacus why?
 
[ SmokeDetector | MS ] Potentially bad asn for hostname in answer (1): Calculating uncertainty in standard deviation by Lior Avrahami on math.SE
 
fp- feedback received
 
[ SmokeDetector | MS ] Potentially bad keyword in body (1): json data storing in mysqli database by ajit on stackoverflow.com
 
@VictorVosMottor Just ignore the message. I would have hoped the ping would be removed once the message was deleted.
 
[ SmokeDetector | MS ] Potentially bad keyword in body (1): How do I link Vertex Groups to Bones? by Vice on blender.SE
 
4:12 PM
Hey @Makeyn ! Why to ignore?
 
fp- feedback received
 
@Undo Ah, got it. Easily done, in the queue behind my retrospective autoflag analysis to amplify my earlier post. It's eye-opening.
 
[ SmokeDetector | MS ] Potentially bad keyword in body, potentially bad keyword in title (2): How will world leaders surive a post nuclear holocaust? by gfhutdf on worldbuilding.SE
 
4:29 PM
Haven't gotten to read it yet, looking forward to getting some time. This is something I've wanted to do/see for a long, long time, thanks for looking at it!
 
@Undo what?
 
[ SmokeDetector | MS ] Potentially bad keyword in body (1): Как сверстать этот блок? (Плотная упаковка) by Dmitry on ru.stackoverflow.com
 
fp- feedback received
 
@VictorVosMottor I'm doing some stats analysis on post scoring and autoflagging.
 
@VictorVosMottor You asked about and linked to a message which was both deleted and moved to a Trash room. Those are really strong indicators that it's time to just let things drop. There's quite a bit of history involved, and rehashing it would, potentially, be disruptive. Given that you were pinged in the message which was deleted, I can understand some confusion and curiosity about the ping. However, the room doesn't need to have us go back over everything to give you the context.
If you're really curious, then the transcript of this room is available to read, but we don't need to rehash things again.
 
4:44 PM
Fun fact: 94 of 193 fp autoflags over a score of 160 were affected by the scoring. Of those 94, 78 would have not flagged with the adjustment.
 
[ metasmoke ] ci/circleci: rails-tests failure on 3dbc4d3: Your tests failed on CircleCI
[ metasmoke ] ci/circleci success on 062136b: Your tests passed on CircleCI!
[ metasmoke ] ci/circleci: rails-tests failure on eb14955: Your tests failed on CircleCI
[ metasmoke ] ci/circleci: rails-tests failure on 062136b: Your tests failed on CircleCI
[ metasmoke ] ci/circleci: rails-tests failure on 062136b: Your tests failed on CircleCI
 
@DanielWiddis And how many autoflagged TP would not have been autoflagged?
 
@Makyen Getting there. :)
 
[ SmokeDetector | MS ] Few unique characters in body, toxic body detected (154): How to access another class from a class's constructor in python? ✏️ by Recon Alpha on stackoverflow.com
 
tpu- feedback received
 
4:54 PM
Whoop, that's supposed to be tp-
 
5:09 PM
[ SmokeDetector | MS ] Link at beginning of answer (33): npm outdated output colour-coding meaning by Andre Gomes Borges Vieira on stackoverflow.com
 
fp- feedback received
 
165 of 298 fp autoflags >= 155 affected. 146 would have not flagged.
36161 of 64505 tp autoflags >= 155 affected. 1985 would have not flagged.
44 of 60 fp autoflags >= 200 affected. 43 would have not flagged.
34355 of 56219 tp autoflags >= 200 affected. 3298 would have not flagged.
 
@DanielWiddis There we go, that's exactly the kind of numbers I wanted
 
Preparing a summary and will update the post within the hour.
@Makyen Note that with improved accuracy, thresholds can be lowered to increase the tp rate to match, so a reduction in tp does not exactly correlate to a reduction in flags. In fact, it can increase it.
 
fp- feedback received
 
[ SmokeDetector | MS ] Few unique characters in body (94): Question not available anymore ✏️ by Gregor Perčič on gaming.SE
 
tp- feedback received
 
[ SmokeDetector | MS ] Potentially bad ns for domain in body (1): How do I know if my microphone is an ASIO sound device? by Wolf Zwiener on superuser.com
 
fp- feedback received
 
5:36 PM
@DanielWiddis How are there more tp autoflagged posts affected which are >=200 (i.e. 3298) than there are tp autoflagged posts affected which are >=155 (i.e. 1985)? I would have expected the >= 200 to be a subset of those >= 155, so the number of >=200 would have to be lower than, not higher than, the number >= 155.
 
[ SmokeDetector | MS ] Few unique characters in body (94): Question not available anymore 2 ✏️ by Gregor Perčič on gaming.SE
 
@Makyen The subset is the 34355<36161 of 56219 < 64505. The "Would not have flagged" deals with the actual score of the post vs. the revised score that would have taken it below the threshold.
 
tp- feedback received
 
[ SmokeDetector | MS ] Bad keyword in link text in answer, potentially bad keyword in answer (82): Correcting grammatical errors versus content in post-graduate coursework submissions in the sciences by James David on academia.SE
 
5:41 PM
tpu- feedback received
 
@Makyen For example a post with a score of 280 adjusted to 190 with the change would not be in the 1985 set but would be in the 3298 set.
 
!!/watch- deepseaatlanticna\.com
 
iBug/Coral: Executing automatic scheduled reboot.
 
PR#4935 ("Spevacus: Watch deepseaatlanticna\.com") opened by SmokeDetector
 
@DanielWiddis Unfortunately, it's still not clear to me what you're saying. I agree with both "34355<36161 of 56219 < 64505". I'm just not seeing 3298<1985.
 
5:43 PM
[ SmokeDetector | MS ] Toxic answer detected, blacklisted user (151): How to drain/exhaust async generator at ~C speed? by Recon Alpha on stackoverflow.com
 
tpu- feedback received
 
@Makyen second reply may be more helpful. A score adjustment 280 -> 190 would not change the result for the 155 threshold, it would still autoflag. But it would change the result for the 200 threshold, it would NOT flag. So the increase is probably posts in the 250-300 range that would not have flagged.
 
teward/Osiris: Executing automatic scheduled reboot.
[ SmokeDetector | MS ] Potentially problematic ns configuration in answer (1): Problems with ffmpeg input filenames when using bash script by Nick Styles on stackoverflow.com
[ SmokeDetector ] SmokeDetector started at rev 3b6d156 (SmokeDetector: Auto watch of viral481\.com by Mast) (running on teward/Osiris, Python 3.6.9)
Restart: API quota is 10584.
 
@SmokeDetector what happened there? how is the user link deleted when I can see the account in the answer?
 
@DanielWiddis Ah... so what you're saying is that of the original set of TP autoflagged posts, there are 56219 which have a total weight of >= 200. Of those, there are 34355 which would have a different total total weight when the criteria you are using is applied. Of those, there are 3298 which would have dropped below a total weight of 200. So, you're talking about a total of 12 different sets.
 
5:54 PM
[ SmokeDetector | MS ] Potentially bad keyword in body (1): How to print to console on windows with Python? by TIIITAN on stackoverflow.com
[ SmokeDetector | MS ] Pattern-matching product name in body (98): Malex python script (where to get?) by Smashy Alts on softwarerecs.SE
 
@Makyen Not sure what you mean 12 different sets... but everything before that is right.
 
@SmokeDetector edited out
 
@SurajRao User link where? On MS?
 
fp- feedback received
 
@DanielWiddis You made 4 statements. Each was about 3 different sets. So, 12 sets in total. But, we're getting pedantic, so never mind that.
 
5:57 PM
Right. Yeah, too much data. Do you want a venn diagram? :)
 
@Makyen MS shows link: stackoverflow.com/users/12398964 while the actual post has adifferent account
 
@DanielWiddis Nah, I got it, but it might have helped. :)
 
Quick summary: autoflag accuracy (tp/allflags) at 155+ changes 99.54% -> 99.76%, almost half the rate of fp. Autoflag accuracy at 200+ changes 99.89% -> 99.97%, about a third the rate of fp.
 
@DanielWiddis OK. So, what happens when people then adjust their autoflagging conditions down to the same % levels based on the new criteria?
 
@Makyen they will get the same total number of fp's at before and a helluva lot more tp.
 
6:03 PM
[ SmokeDetector | MS ] Link at beginning of body (45): Heremap 6.2/geocode.json returns wrong state and City by Sajitha Anish on stackoverflow.com
 
fp- feedback received
 
@Makyen For example, say my current threshold is 200, I get 60 fp's historically, a 99.89% rate. Under new change at same threshold I reduce my fp's to only 17 a 99.97% rate. I lower my threshold to 193, getting back my 99.89% rate, which gives me 62 fp's, roughly the same as before. (...continuing...)
Before the change I had 56219 tps. After the change it was reduced by 3298 to 52921. But lowering my threshold to 193 gives me 56644 tp's, more than I had before for the same number of fps.
 
@SurajRao That is... strange, but the strange part appears to be on SE's side, not MS.
 
Yes.. the only thing I can think of is account merge... Just a quess
 
@SurajRao That's basically what I also figured had to have happened, based on the publicly available data.
 
6:18 PM
[ SmokeDetector | MS ] Potentially bad keyword in answer (1): Momentjs changes the timezone value by Adebayo Omolumo on stackoverflow.com
 
fp- feedback received
 
[ SmokeDetector | MS ] Offensive answer detected, toxic answer detected (158): Last problem in relation to the logic of linear algebra. by DAstro on math.SE
 
tpu- feedback received
 
totally not work slack
 
I've definitely never typed commands intended for a terminal into a chat window
usually it's one of my weird inexplicable aliases, too, so I have to explain myself when they wonder "where did you mean to type just 'ts'?" :-p
(it's short for tig status, if you're curious) Another weird one is gcane, which is short for git commit --amend --no-edit
 
@SmokeDetector why
 
@RyanM Blacklisted user - blacklisted for //math.stackexchange.com/a/3840311 (metasmoke.erwaysoftware.com/post/266463) by the metasmoke API
 
fp- feedback received
 
@DanielWiddis But, that doesn't make sense. If the instantaneously calculated %TP is the same, then you'd have the same proportional number of TP to FP, not lots more TP for the same # FP. A new method of calculating total weight might result in more accurate predictions of future events, but if you're picking a set %TP (not a set weight) and then looking at the ratio of TP to FP that you get in the existing dataset, then that ratio isn't going to change, because %TP describes that ratio.
In other words, 90% of 100 is 90 and 90% of 200 is 180. The ratios don't change to give lots more TP for fewer FP while maintaining the same percentage.
 
6:39 PM
[ SmokeDetector | MS ] Manually reported answer (94): Last problem in relation to the logic of linear algebra. by DAstro on math.SE
 
tpu- feedback received
 
@SmokeDetector User is continuing their bad behavior from metasmoke.erwaysoftware.com/post/266463
 
@NobodyNada User is blacklisted (828282 on math.stackexchange.com).
 
tpu- feedback received
 
6:41 PM
they have some legitimate posts, but I'd rather keep them on the blacklist for now
 
Teams post updated with new data on autoflag accuracy.
 
[ SmokeDetector | MS ] Offensive answer detected, potentially bad keyword in answer, toxic answer detected, blacklisted user (230): Last problem in relation to the logic of linear algebra. by DAstro on math.SE
 
tpu- feedback received
 
@Makyen ... misread the question, adjusting answer.... ignore
 
6:56 PM
[ SmokeDetector | MS ] Potentially bad asn for hostname in answer (1): how to design the footer with svg image that has waves style? by YlmRdm on stackoverflow.com
[ SmokeDetector | MS ] Few unique characters in answer (87): Two-weeks' notice ✏️ by LPH on english.SE
 
tp- feedback received
[ metasmoke ] ci/circleci success on 44af36c: Your tests passed on CircleCI!
 
[ SmokeDetector | MS ] Potentially bad keyword in answer (1): Is Superman always holding back? ✏️ by Shadows Edge on scifi.SE
[ SmokeDetector | MS ] Link at beginning of body (45): Necromancy subclass balance suggestion? by Nachtkalmar on rpg.SE
[ SmokeDetector | MS ] Potentially bad keyword in body (1): Can I use Gluon Mobile free version to deploy commercial solutions (software)? by MegaBoost on stackoverflow.com
 
7:22 PM
@Makyen There were 56,279 autoflagged posts at 200 or higher. Existing scoring fp rate is 60, a 0.1066% rate. There were 59,393 autoflagged posts at 193 or higher. Under new scoring that reduces to 56,706 posts. Apply the identical 0.1066% rate to the larger number of posts available to flag, 56706, and you get 60.46 fp + 56645.54 tp. For the same total fp you have 426.54 more tp.
 
@SmokeDetector fp- but off-topic
 
[ SmokeDetector | MS ] Potentially bad ip for hostname in answer (1): What is the theme of Glasslip anime? by RadiantArrow on anime.SE
[ SmokeDetector | MS ] Potentially bad ip for hostname in answer (1): What are the words and meaning that compose the title of Glasslip? by RadiantArrow on anime.SE
 
Another example. I'm required to have 99.75%, best I can do is 193 threshold with 58,842 available posts. I miss 105 at a 0.178% fp rate.
Now, revised scoring, I can keep that same fp rate by lowering threshold to 174. I get 920 more posts to flag, and miss 106 of 59,672 (0.177%). At the cost of 1 fp, I've tp'd an additional 919 posts. But say you don't want fp to increase at all... fine. Use 175 threshold. I get 59,676 posts to rate, missing 103 of them (0.173%). I have reduced my fp by 2, but still increased my tp by 836.
 
7:41 PM
[ SmokeDetector | MS ] Potentially bad ip for hostname in body (1): Is there a way of findning the mistake, in which row there is a breakpoint by Marcelo on stackoverflow.com
 
@DanielWiddis That's not the same #FP. The extra TP are what you get when you have the extra 0.46 FP (60.46 FP instead of 60 FP) at the 0.1066% rate. But, really, I think we've gone down a rat-hole here.
 
@Makyen My second example adjusts for your objection. The problem with your 100 to 200 example is you assume the same % applies as the total #s increase, but it's really the accuracy of the marginal difference that matters.
@Makyen But assuming exactly the same % score, yes, there'd be a small increase in fp for a comparatively much larger increase in tp.
 
@DanielWiddis I'm not saying that I don't want FP to increase at all. I'm saying that if you have total posts = #TP + #FP, then with a fixed %TP, the ratio of #FP to #TP isn't going to change, because the %TP describes that ratio.
 
@Makyen Fair. I'm also saying with a tiny ratio the concern fades into roundoff.
As Spock would say, "Logic clearly dicates that the good of the many additional spam posts autoflagged outweigh the needs of the few wrongly flagged posts."
 
[ SmokeDetector | MS ] Potentially bad keyword in answer (1): Telegram wrong URL host by Milawd heidari on stackoverflow.com
[ SmokeDetector | MS ] Offensive answer detected, potentially bad keyword in answer, toxic answer detected (159): Automating Alchemy pad segments in Logic Pro X by Random logic user on music.SE
 
8:04 PM
tpu- feedback received
 
Makyen/EC2-linux: Executing automatic scheduled reboot.
 
8:27 PM
[ SmokeDetector | MS ] Potentially bad keyword in answer (1): Run PHP script constantly as windows service by ssarljames on stackoverflow.com
 
fp- feedback received
 
[ SmokeDetector | MS ] Potentially bad keyword in body (1): Home network DSL/WAN question by Alexandru Cumpanici on networkengineering.SE
 
8:44 PM
@DanielWiddis Ah... but there won't, necessarily, be more posts autoflagged. If the %TP which is used remains the same, then the same number of posts will be flagged. It will be somewhat different posts, but it will be the same number, within the errors introduced by rounding. What we're hoping for is that there will be more TP posts autoflagged and fewer FP posts autoflagged (i.e. that the predictions will be more accurate).
In other words, that the predictions of the probability a post will be TP, based on using the derived criteria/metrics, will more accurately match the historical %TP which is obtained when those criteria/metrics are applied to the historical data. (i.e. What we're actually talking about is how good the new "weight" number is at predicting if a new post will be TP or FP.)
Both methodologies are equally accurate when looking at past performance and picking a %TP, because %TP is a fixed number and doesn't change when looking at static past data. What we're really wanting to know is if a new way of calculating some other metric, in this case probably both called "weight", is a more accurate predictor of future %TP. In other words, when new data comes in, does the metric we're using change much wrt. it's equivalence to %TP as a result of including the new data.
That's a hard thing to know or accomplish. It's made harder by the fact that the detections which are used to generate the "weight" have changed over time. The data available upon which predictions are being based, is all based on what the detections were at the time each specific post was detected, not as the detections are at this time (i.e. the detections change over time; both individual detections change and detections are added and removed).
In other words, if the current state of the detections was used to re-scan every post in our database, then many of those posts would have different detections triggered, which would result in different weights. There would be posts which would get fewer, or no, detections (i.e would actually, now, score a lower weight), and posts which would get more detections (i.e. would actually, now, score a higher weight).
There are, of course, also a considerable number of posts that are not in our database which were scanned and not detected, but would now be detected with the growing/changing detections. How many is unknowable, without a complete record of all posts, even including things which SE doesn't record (e.g. grace period edits).
In other words, the data upon which all of our historical "weight" numbers are based does not reflect the current state of the detections. The further back in time you go in the data, the further it diverges from what the data would be, if each post was re-scanned with the detections as they currently are.
The analysis you're doing is quite helpful. I'm hopeful it will result in more accurate predictions of what will be TP or FP (i.e. how well a derived metric (e.g. "weight") will match actual).
 
[ SmokeDetector | MS ] Mostly punctuation marks in answer (36): Swapping 6 queens in a 4x4 grid by Albert.Lang on puzzling.SE
 
fp- feedback received
 
9:06 PM
[ SmokeDetector | MS ] Blacklisted website in answer, potentially bad keyword in answer (94): SharePoint 2016 FBA on-prem DUO .NET WebSDK by Martin on stackoverflow.com
 
fp- feedback received
 
@Makyen I understand what you're saying here, but I think it's slightly more complicated. Consider Type I error (falsely flagging non-spam) and Type II error (not flagging true spam). For any given "what should the real score be" set of posts, we are reducing the overall variance of our score. For the same sample size (all posts with a score >= 1) if we keep Type I error the same (by allowing lower thresholds) we will case Type II error to decrease.
 
[ SmokeDetector | MS ] Link at beginning of answer, potentially bad ns for domain in answer (34): Crystal Reports newer runtime broke old applications by william haynes on stackoverflow.com
[ SmokeDetector | MS ] Luncheon meat detected (41): Resolv.h remove UDP/TCP separator by The TeenBoss on stackoverflow.com
 
9:24 PM
well lucky us, the backup power supplies are working lol
(random power outage)
 
fp- feedback received
 
[ SmokeDetector | MS ] Bad keyword in body, potentially bad keyword in body, potentially bad keyword in title, blacklisted user (168): {@}}+27719247950 BUY 100% COUNTERFEIT MONEY & QUALITY DOCUMENTS +256789125443 by pricess anith on money.SE
 
tpu- feedback received
 
@Undo Here's your temporal analysis. Now to take the rest of Friday off. :)
 
@ThomasWard That's never fun. I'm glad to hear the UPSs are working.
 
9:32 PM
@Makyen yup.
they were running for maybe 8 minutes before the power came back so
 
fp- feedback received
 
10:30 PM
API quota rolled over with 8208 requests remaining. Current quota: 19999.
math: 487
stackoverflow: 306
physics: 88
superuser: 84
askubuntu: 75
electronics: 63
unix: 58
salesforce: 50
serverfault: 44
mathoverflow.net: 43
english: 39
diy: 30
ell: 29
worldbuilding: 28
dba: 28
security: 27
meta: 26
apple: 24
scifi: 23
rpg: 21
chemistry: 20
es.stackoverflow: 20
quant: 20
codegolf: 19
money: 18
ai: 17
drupal: 17
politics: 17
gaming: 16
christianity: 16
music: 16
wordpress: 15
puzzling: 14
datascience: 14
graphicdesign: 14
ethereum: 14
dsp: 13
ru.stackoverflow: 12
softwareengineering: 12
[ SmokeDetector | MS ] Bad keyword in answer, potentially bad keyword in answer (87): Want to try keto diet but have no money by Rwan on fitness.SE
 
tpu- feedback received
 
[ SmokeDetector | MS ] Bad keyword in answer, potentially bad keyword in answer, blacklisted user (158): weight loss/diet plan help by Rwan on fitness.SE
 
tpu- feedback received
 
Random thought. If a post isn't autoflagged, but we have enough users in this chatroom tp it, could it not be re-processed through autoflagging with "higher weight"? For example the two Keto posts above are clearly spam but are awaiting more flaggers... which we have tons of waiting in the wings.
 
10:37 PM
[ SmokeDetector | MS ] Bad keyword in answer, potentially bad keyword in answer, blacklisted user (158): Strict keto diet - still not losing weight by Rwan on fitness.SE
 
tpu- feedback received
 
4 tp votes on each one, still on the site...
 
[ SmokeDetector | MS ] Offensive answer detected, potentially bad keyword in answer, toxic answer detected (159): How to trace origin of script injected in my webpage by user113965 on webmasters.SE
 
tpu- feedback received
 
@SmokeDetector appears to be an attempt at exploiting the site, they've edited the post to have more obnoxious formatting too
 
10:43 PM
[ SmokeDetector | MS ] Potentially bad keyword in body (1): What is the fight between the Sanzer and Rizhin Rebbes? by Margot Jaffe on judaism.SE
 
@DanielWiddis Yeah, this is something I've been curious about as well. I always wondered why TP's didn't start autoflag engagement. But, then again, a tp response doesn't always mean spam.
 
@Spevacus true. Hadn't thought of that.
 
It'd be nice if we could manually incur a flag from the SmokeDetector account. Plenty of spam posts have sat at -5 needing only one last spam flag to get nuked.
 
Similarly, I wonder how much we'd reduce autoflag fp by just waiting for one tpu feedback before sending the autoflags.
 
But I can understand why not. If it was cast in error, there's no way to retract it.
 
10:49 PM
There are a lot of us here, and we're usually pretty quick to hit the autoflagged posts.
Might be less effective during times/days when a lot of us are away, though. But at least we're spread around the world enough that time zones aren't a major issue.
 
And I guess nobody's really harmed by an extra 15 minutes of keto advertisements. Posts are still there, btw. ;)
I guess even one flag is enough if/when a moderator gets to it.
curious if the mod spam flag queue is prioritized to put multi-flagged posts at the top
 
If anything, it does increase visibility in general. If there are multiple spam flags, and they're picked somewhat randomly, the chance for that post to come up next in the queue is increased because of the extra flags.
I know that red flags are prioritized in their queue. Dunno about how it handles multiple flags.
 

« first day (2504 days earlier)      last day (1306 days later) »