wait, 90?

Wtf?

:O

Huh, 0/1/2 is apparently a condition above 99.75%

....

no it isn't

oh bugger

it's these people playing with the sites

this is what people are talking about when they worry about people gamifying autoflagging

I thought there were limits on new users?

SU, AU, Gaming, AD, Drupal, TWP, Astronomy

@ArtOfCode figures

drupal and astronomy is probably what throws it off

then the rest is whatever they could fit in

Hang on, why would autoflagging come into this? Are you saying that anyone with a metasmoke account can come here, mark something as a tp for smokey and that will cause an autoflag to be cast?

no no no

has nothing to do with that

basically what's going on is that you can set specific conditions for certain sites

@quartata Phew. That did sound weird :)

and people have found groups of sites where any two reasons is enough to be 99.75% accurate

Argh. Bugger indeed.

that will get them lots of flags

But still above the min accuracy threshold?

I mean, it's not like they're getting around the accuracy threshold or anything

yeah

Which would suggest the chances of fps are still ridiculously low

I'm in two minds about that. On one hand, it does feel like gaming the system. On the other hand, it is above 99.75%, so that should be enough.

It's just that they're trying to maximize their returns in a way

yeah I'm not talking about it being a risk at all

32579/72 tp/fp

wow, that is really good for only those 7 sites

On? Those sites with those rules?

my condition which is across all sites only catches 46K

@SmokeDetector Laughs that it's obviously just me and @iBug...

@terdon yup, those 7 sites with min weight 0, max rep 1, min reason count 2, gets you 99.78% accuracy with those counts

@ArtOfCode christ

Max rep 1 of the poster, presumably?

Yes

Seems like there should be a global minimum weight?

reason weight/number of reasons/max reputation

righto

it's just

I would prefer that conditions work anywhere is the thing

you know?

but eh it's probably not a giant problem

72 fps is, like, really good. I'm 46688/114 and I'm 184/1/2 everywhere

if people are doing this kind of specialization and it ends up with less FPs then I guess that's a good thing

@quartata Yeah, but it makes sense that each site is different.I would expect some to need less for that level of accuracy and others to need more.

I mean if it's only (99.75% of the time) flagging tp, it shouldn't be a problem?

@WELZ it's not. the problem is more philosophical

i.e. do we tolerate it, even though people will think we're optimizing for something that we shouldn't be optimizing for (how do I get all the flags)

Make it 8 sites, GD can also be added

Team launches at 1100 Eastern tomorrow.

@ArtOfCode nice

that is, I get access to it at that point... y'all probably won't have access til the end of the day

@WELZ just to clarify: you're not in trouble, it's just food for thought

I think we should be asking ourselves these questions so that we're not taken off guard when others do

Don't forget to add in all the sites to your Flagging Preferences (new sites aren't automatically added)

[ SmokeDetector | MS ] Potentially bad keyword in answer: How to restore my old deleted WhatsApp messages by roseline on android.SE (@iBug)

tpu- by Floern

@quartata The only time I wouldn't tolerate it is if someone was using multiple conditions to sidestep the 99.xx% accuracy number.

@Undo like, overall. yeah.

hadn't considered that.

(I.e. 99.8 percent on 1, 99.8% on the other... But the fps are completely exclusive and you actually have 99.6% overall.)

@Undo speaking of... wanna check these?

@ArtOfCode Didn't I do that already? Thought it came out to .01% low, which is meh

think they've changed

automatic validation process on that wouldn't hurt, either

!!/watch (\+1)?518\W{0,3}504\W{0,3}9376

@K.Dᴀᴠɪs You don't have code privileges, but I've created PR#1687 for you.

@ArtOfCode It's a pretty heavy process to do automatically, but yes.

@Undo s/automatic/automatic on-demand/

stick a button on that page I linked to make MS do the validation, just cuz it's easier than me remembering the queries for it

Works

If someone approves that PR, I don't think we need the 1st group in there... Just 518\W{0,3}504\W{0,3}9376

^^^ That phone number has come up three times before. All TPs.

So I've come to the conclusion that there's no real point in stacking Bottle on top of Tornado when I already have Tornado experience and I need websockets which means a cludgy fallback handler + WSGI

"there's no real point in stacking Bottle on top of Tornado" -- today in quotes to take out of context

[ SmokeDetector | MS ] Blacklisted user: Could hacker change data in one smart contract with out call contract function? by Yudao Yan on stackoverflow.com

@Undo I am looking at the worst autoflag stats. They're still pretty good mostly, but in order to draw robust conclusions, I would also need the total number of posts processed. Is there an easy way to get those?

This is what I have so far:

Site Name                   FP  TP    PPV (TP/total)
Arqade                      3   285   0.982759
Stack Overflow              17  1379  0.982194
Worldbuilding               0   49    0.98
Personal Finance & Money    4   231   0.978814
Travel                      2   91    0.978495
Science Fiction & Fantasy   1   43    0.977273
Mathematics                 1   40    0.97561
Academia                    1   38    0.974359
Android Enthusiasts         2   75    0.974026
Bitcoin                     2   37    0.973684

fp- by Floern

@SmokeDetector who blacklisted this guy?

sd - why

The PPV is useful, but what's really needed is specificity and for that we need the number of true negatives (so posts that weren't spam and weren't autoflagged).

Or, at least, the number of posts to have an idea of the sample size and get some proper stats into it.

Looks like the post's been reported twice: metasmoke.erwaysoftware.com/post/109298

fp by NobodyNada

@SmokeDetector fp

@NobodyNada Registered question as false positive and removed user from the blacklist.

Looks like it was self-retracted feedback

¯_(ツ)_/¯ some fgitw

@terdon site dashboard?

Oh true negatives. That's.... difficult

Yeah, true negatives will probably be impossible.

But at least the total number of posts processed would help.

I mean the trouble is more false negatives

You can pretty safely assume we scan every post