Topic: Quickseller escrowing for himself - page 34. (Read 33682 times)

So much drama!

Apart from his username, I have no idea.

So now QuickSeller got banned in this forum?

 

I verified the PGP signature.  


Edit: Meant for QS only.
(Someday, you got to admit that you are wrong with at least two things.
One is that tspacepilot is not a scammer.
The other is that dadice is not a scammer. (slightly off-topic, sorry).

I have no idea how many others got affected.

But, I am grateful that you accepted that I am not a scammer and removed the negative trust. I still think I got lucky that day that you found at least considered one of them as valid though the original scammer could have easily covered that part too.)

It would probably break on sold accounts as there would be so much old counter information muddying the new poster's style. Unless you manually cut it off at the proposed sale date and analysed the two sets separately.

I suspect that Panthers is an alt of QS too. I am very sure, either that or Panthers have precisely the same attitude, the skepticism, and everything I relate to QS, and he seems to be taking his side almost everywhere.




Check QS's personal text.



Edit2: If someone covered up the left hand side of the post and asked me who it was I would say it is QS.
https://bitcointalksearch.org/topic/m.12323666

Except that he's right here.  Even he's not denying it.  He's just sorta posting off-topic nonsense in order distract.

I read almost the whole thread, but didn't see "forced break", what is that?
Sorry about the dumb question.

Good question.


About as promising as the default trust list, that neither of you like. See my updated post above

Does forced break = 3 day ban?

Someone would have brought this up sooner or later, so I guess I will now and you can get it out of the way. https://bitcointalksearch.org/user/quickseller-358020 - "Forced 3 day break from the forum Be back Saturday". This thread was created today.

BTW, a scammer could and will most likely use multiple posting styles. Your algo could make other members believe they are dealing with someone with no alts. The algo could also be wrong and flag someone as a scammer when they just post similar to a scammer, or that scammer has studied their posting history and decided to copy it. A scammer could claim the same : Setup an alt, post normally, scam and when caught say someone copied him. Unlikely but over time possible. This could result in either a scammer getting away or the one not scamming having his reputation ruined. What a minute, that kind of reminds me of the flaws in the default trust list a few members keep bringing him.

Once again, not taking sides here. I'm just pointing out the flaws. The algo is pretty cool but appears to need a lot more work and even then could be wrong, resulting in someone's reputation getting ruined or someone getting away. Same as now. So nothing really has changed.

That doesn't make any sense to do.  This account has 13 posts.  It's nowhere near enough data to get a coherent sample.


People can draw their own conclusions.  I think that comparing the quickseller model as a predictor of three classes was pretty informative.  The three classes are (1) a known alt; (2) a suspected alt; (3) people we know aren't his alt.  The data I presented show that the suspected alt and the known alt are predicted equally well and much better than the people we know aren't his alt.  They other models provide context, predicting the posts of another person with a model that isn't theirs you end up with perplexity from 250--400.  When we used quickseller's model to predict the text of his alts, we got perplexity around 100-150.

We could tweak the model parameters and we could run more experments.  I'm happy to help someone else to do that but I don't have the time or resources to do nothing but run models all day.  What's more, it seems that QS has given up even trying to defend himself here.  If panthers52 wasn't his alt, wouldn't the first thing he'd do be to say "nope".  Instead, he says this:

^^^ Still going.  Writing weird nonsense about some post or thread which is intended to distract and cause confusion.  If you can't handle the message, attack the messenger.  You're looking more and more desperate with these sort of shenanigans.  I do sorta love how as his emotional level is going through the roof, his attempts to disguise his demeanor as quickseller have sorta fallen away completely.

Quickseller, more on-topic here would be to confirm or deny that panthers52 is your alt.

Apparantely, it might not even be an issue:


I don't trade here so I don't know what's considered honest/dishonest.  My intuition is that suggesting to use yourself as an escrow is dishonest, but I leave it to the community to decide.

I know tsp from when he created a useless threads whose point is little more then a better-then-most-veiled attempt to boost his post count. The thread was about him claiming to have a transaction that would not confirm for days, I tried calling him out on this being BS, and he eventually gave me a5e169d60a797e4585c299cfa8bd2aff457f4d80a5b0c70e0f467e35fd21e1ad which was broadcast ~an hour prior to it confirming, and confirming in a block that was found days after the thread was created I am fairly confident this was a transaction to himself as cover for creating the thread, however I didn't think I had strong enough evidence to create a meta thread regarding the pointless thread. For some strange reason, tsp thought that he wanted me to post in his thread about QS and as soon as I did he claimed that I was QS and started to flame me.

The only reason I can think that tsp wanted me to post his thread about QS is because I had disagreed with him wanting to make the trust system changed so it would be easier to scam and easier to farm trust.

From what I can see, tsp is going to troll anyone for months who disagrees with him, and as a result I would suggest anyone reading this to avoid giving any reason to say that tsp is wrong in this thread or any other one. If you do disagree with tsp then prepare to be trolled and harassed for months by him, including from a potential sock-puppet.

It would be a good baseline to start with. Instead of attempting to hunt and search for proof that might not even exist, it would at least provide someone with something to work off of. It would at least show that it is possible for two accounts to be alts for further research instead of going directly to searching for blockchain evidence which can take a very long time. Such a program would also be useful for seeing whether an account was sold and roughly when.

Great idea! Then people can do to others what tspacepilot claims quickseller did to him, make accusations without solid proof, just assumptions based on what others say or in this case, a script / algorithm or whatever you want to call it. No?

Not taking sides here. Would I be shocked that quickseller escrowed for himself? No. I've seen other high ranked members do it.

It might.  But it only becomes useful when you have a lot of text to start out with.  In QS' case, he's written a whole lotta words .5 million words is not a small amount of text.  And it's especially useful here because we have a known alt to compare the model's accuracy with.  That is, it's quite interesting that the model predicts acctseller's text with the same metric as panthers52.  I imagine that there are few cases where the person suspected of scamming has written as much as QS.  But you're right, it could be useful.


I think the wikipedia article on perplexity is a reasonably good place to start https://en.wikipedia.org/wiki/Perplexity.  I'm not a very good teacher, and much better prose than I can produce has been written to explain the metric.  I think the simple way to think of it is as the cross-entropy of the model<->test-corpus relationship.

Here's another web-page which talks about it and how it's used to predict text. http://itl.nist.gov/iad/mig/publications/proceedings/darpa98/html/lm30/lm30.htm


It should be more or less linear.


You could certainly compute pearsons r on the matrix I provided.

I also have the intuition that a t-test could be relevant.  But it's not clear to me at the moment how to set up the parameters.  Perhaps someone smarter than me will pick this up and/or describe how the t-test would work in this particular case.


As I said above, I doubt it would become generally useful unless the people in question have a large number of posts.  Having a lot of data to start from makes a better model.  You couldn't use this on newb accounts.

You should definitely write this into a program and put it in services or project development. It would be great for analysis on finding people's alts, not just quickseller. In fact, maybe quickseller himself might even use it.

I'd say that's a reasonably good layman's terms summary of what's going on here.  But to be more concrete, what it shows is that a large-ish statisical model of quickseller's posts predicts the posts of acctseller and panthers52 equally well and signficantly better than it predicts the posts of me, dooglus, or hillarious.  The reason it's important is that these kinds of relationships are "hidden in plain sight" in the language we use all the time, they're not easily maniuplated.  They're built up over the 1/2 million words that qs has posted on this forum.  The statistical relationships in that corpus allow us to look for matches against other accounts.  I've posted al the code and techniques here for anyone who wants to replicate the experiment or try the techniques on other accounts or pairs of known alts.

It's also important to note that this isn't the only evidence that Panthers52 is QS's alt.  There's actually a sorta overwhelming amount of circumstantial evidence.  I suppose that sooner or later we'll hear from badbear about whether there's any IP evidence.  There's also a particular "tick" that QS and his alts have, and as far as I've seen, no other accounts have this.  I won't say what it is yet because it's nicer to leave it hidden in case QS creates more alts.

My issues with Quickseller have to do with the fact that he's been bullying me relentlessly for nearly a half a year.  I didn't expect that I would uncover evidence of him doing an escrow scam.  This came out on accident, as he pulled in another sockpuppet account to try to attack me with.  I leave it to those who trade with him and use him as escrow to decide whether it's dishonest to trade with someone who's an alt of the escrow provider.

I think this is really interesting analysis that could help with the issue of sock puppets in the forums if it achieved legitimacy here.

I have a remedial question.

For the index numbers why is lower better?  It seems like you are using one users posts' content to predict the others, right? Or amount of commonality?  I tend to conceptualize this as a percent or ratio, how does that translate into the index numbers you've calculated?

What do the number mean in relation to each other, are they a linear index?

Yamato has a point right?  All your analysis does is say their posts are written in a similar manner.  Can you do some statistical test to show how likely that is just random chance? e.g. a t-test?

Because I follow dooglus's posts I have been keeping up with your other thread in meta.  I initially didn't think panther was an alt but when he failed to follow up on this:


I started to have my doubts.

If this type of analysis does become a useful tool on the forum I can see lots of applications, for example in detecting account sales.