Ladder Site Online...

[Insinuator]

On a different topic entirely, I just went to the Ladder homepage and got this error all over the screen:

Code: Select all

Warning: mysql_connect() [function.mysql-connect]: Can't connect to local MySQL server through socket '/var/run/mysqld/mysqld.sock' (2) in /home/subversiva/data/www/ladder.subversiva.org/conf/variables.php

[SalsaRocoto]

Hey all !

looks like there has been some action going on on this thread seems like I opened the Pandora box MUHAHAHA

peace and love section

On a more serious tone, I am really sad when things get a bit too aggressive guys. Remember we are just debating about a game, considering a rule change to the ladder. No need of insults, no need to ask to have your account deleted either: we are in the process of making a decision. Let's take it all together and if you really are not happy with it, sure you can leave forever.

nani

Secondly, I join my voice to all the players that asked to have nani and his aliases unbanned. IMO this ban should never have happened

Rigor

Thirdly remember that Rigor is a great ladder admin. one year ago 70% of the games were played on 4 maps only. Look at what we have now! this has been the work of many, but Rigor sure has been very involved in that great change. Rigor doesn't get anything for all the time and energy he invests in administrating the ladder. Try to remember that before criticizing him too much please.

Now to the debate ! basically the two points I have taken out of all this is:
* top players having multiple aliases provide more opportunities to play games with top players. that is GOOD
* when an alias is created until it reaches its stable level the opponents are hurt because they dont get the points they deserve if they win and they loose more points than deserved if they loose. that is BAD.

Many players have said this second point doesn't really matter because the hurt is not big and not for long. Values like 10-15 games have been given. At that point I wondered is it really true ? I thought we needed hard facts, a serious quantification of how hurt are the opponents of an alias.

So I did my home work and figured out the parameters of the ELO system used by the ladder. Then I took Demogorgon as an example and calculated the cost for every opponent of Demo assuming Demo would have had a rating of 2260 from the beginning. In other word if you played Demo when his rating was 1600 and lost, I recalculated how many points less you would have lost if he had been 2260. I did that for every game, and here is the result:

Code: Select all

Total cost   cost for winners  cost for loosers
    449            105             344

And here is the full list by pseudo:

List

Code: Select all

Player      Total ELO pts lost     for winning     for loosing
Neki                33                  9              24
Henrythe12th        30                  6              24
c4rLOs8             30                 19              11
Oook                24                  8              16
Skb                 17                  2              15
Hellboy             16                  7               9
Dreadnough          14                  4              10
Nelson              13                  7               6
The_Black_Sword     13                 10               3
Josh-The-Boss       12                  0              12
Thefish             11                  0              11
ParadiseCity        10                  2               8
Tesafilmchen        10                  0              10
gaute1              10                  2               8
Cremember            9                  4               5
Hover                9                  0               9
Huumy                9                  1               8
ArtemisX             8                  0               8
stop_chemtrails      7                  0               7
tekelili             7                  0               7
Ichorid              7                  0               7
Yzr                  6                  1               5
Aca                  6                  0               6
Frosti               6                  0               6
Quetzalcoatl         6                  0               6
Fantri               5                  5               0
Jelly_the_Mad_Doggy  5                  0               5
Kira1                5                  3               2
khiM                 5                  2               3
Fosprey              5                  0               5
Silent               5                  0               5
Others              96                 13              83
Grand Total        449                105             344

So for example, neki would have 33 more ELO if demo had been 2260 all the time.

unfortunately now I am too tired to discuss those results, but I will in the next days post my Excel file so everyone can play with it, post the ladder Elo parameters and suggest some ways we could change some parameters without altering any code.

Salsa goes to bed

[Scatha]

SalsaRocoto:

The effects are both more and less than your calculation suggests, but mostly less. Given that neki was 33 points down from where he "should" have been, in his next few games he would tend to gain more points / lose fewer than otherwise, and in the long term his rating should stabilise to the same value in either case. I'm not sure how many games are needed to effect this change.

On the other hand, neki would have been underrated, so his opponents would have had the same problem recreated for them in miniature. Fortunately this is exactly the kind of thing ELO is set up to deal with. I'll note that in gaining ~750 rating Demogorgon only cost other players ~450; the total discrepancy introduced to the ladder is decaying. It's this decay which means that people don't worry too much about the effect here (but it would be nice to know really how fast it decays -- I'd guess it approximates as exponential decay, but I don't have a proof!).

Edit: Thanks for doing that analysis, though!

--

To people discussing the merits of allowing versus banning multiple accounts at this stage: I think if they were banned we would have some of the current players who have had multiple accounts give them up and henceforth play with just one account. However we have a pretty good historical precedent of players who are generally enjoyable opponents independently and secretly making alternate accounts when it's against the rules; I don't think there's any reason to believe that this wouldn't happen in the future (with perhaps a new set of such players).

On the other hand, if there are going to be people with multiple accounts, just about everyone agrees that it would be better if those were public. I think lots of people who have them would be happy to make them public if it were allowed. This seems like a win-win situation.

Some people may regard the following as an added bonus: it would make it far less defensible to have a secret second account, so if such ever came to light you might be able to avoid the current controversy.

[Madlok]

What happened to the ladder site? Admin decided to try change the FAQ? I hope that everything is under control.

[Insinuator]

On the other hand, if there are going to be people with multiple accounts, just about everyone agrees that it would be better if those were public. I think lots of people who have them would be happy to make them public if it were allowed. This seems like a win-win situation.

Not everyone would, though. You can't just rely on people's "good will". We saw how that failed already. It seems to me there are only two real choices here:

1) Change the rule to allow multiple accounts. Make it explicit so that new people joining know they may be playing against several accounts that are the same person. People don't have to reveal their aliases, it will just be known that they might have them.

2) Enforce the single account rule through some technical process such as screening for multiple accounts through IP traces.

In my opinion, solution 1 sucks but is the easiest. Solution 2 would require a skilled person to do significant work to implement it, so it is highly unlikely to happen. But other solutions all depend on players relying on other players to "confess" their aliases, which is a naive feeling and not a practical solution.

[Scatha]

OK, but how about:

1') Allow multiple accounts, but encourage people to make multiple accounts public, and provide an easy mechanism for them to do so.

It doesn't differ from 1) in any strict or enforcible manner, and certainly you can still evade the system if you want to (just like currently), but it changes the tone of the thing, and I think at least a reasonable proportion of people who want multiple accounts would be happy to comply with this, and that the people playing against them would be happier for the openness.

[Madlok]

2) Enforce the single account rule through some technical process such as screening for multiple accounts through IP traces.

IP can bee changed. And what with players on a local network or public WiFi?
In my country ADSL is very popular. Every time you connect to the internet you get different IP.

[Caphriel]

2) Enforce the single account rule through some technical process such as screening for multiple accounts through IP traces.

This is not technically possible. It might work for people whose ISPs assign static IP addresses, but for others, their IP address changes on a daily basis Everyone in the same subnet will appear to be the same "person" after enough IP assignments. Also, multiple players on the same connection, etc. As the worst-case scenario, no IP-based or similar system can distinguish between two players using the same computer. Given the current state of the technical side of the ladder to my knowledge, implementing a technical method of identifying multiple accounts belonging to a single user is impossible.

Ninja'd, but posting anyway for more detail.

[Insinuator]

1') Allow multiple accounts, but encourage people to make multiple accounts public, and provide an easy mechanism for them to do so.

It doesn't differ from 1) in any strict or enforcible manner, and certainly you can still evade the system if you want to (just like currently), but it changes the tone of the thing, and I think at least a reasonable proportion of people who want multiple accounts would be happy to comply with this, and that the people playing against them would be happier for the openness.

I see where you're coming from, but think about this: Say most people are open about having multiple accounts. Now there are others who would prefer to play with a single account (Myself included. I still think multiple accounts is, at best, frivolous.). But now think of a third group of people who would continue to hide their multiple accounts so that they can use them to obviously cheat, i.e. play against themselves to inflate their own elo. I don't think people will stop doing that just because others are open with their mutiple accounts. I especially feel that way considering how long it took for the people in this thread to even admit they had them, even though they claim to have done so for "benevolent" reasons.

This is not technically possible. It might work for people whose ISPs assign static IP addresses, but for others, their IP address changes on a daily basis Everyone in the same subnet will appear to be the same "person" after enough IP assignments. Also, multiple players on the same connection, etc. As the worst-case scenario, no IP-based or similar system can distinguish between two players using the same computer. Given the current state of the technical side of the ladder to my knowledge, implementing a technical method of identifying multiple accounts belonging to a single user is impossible.

True, yes. But couldn't it be possible to associate the IP address of the user with his account name each time he logged on? For instance, say Insinuator logs on as 172.65.14.22 on Monday. The site logs that address. Now, say Insinuator tries to log on an hour later as Caphriel. Assuming Insinuator hasn't reset his Internet connection, his IP address should still be 172.65.14.22. The site would compare that with the last log-in from that address and block it. Someone could get around this, of course, but it would provide a level of security.

And, of course, there is no way to distinguish between two people on the same computer. That's a little obvious. Unless we ask Big Brother to step in...

[Scatha]

1') Allow multiple accounts, but encourage people to make multiple accounts public, and provide an easy mechanism for them to do so.

It doesn't differ from 1) in any strict or enforcible manner, and certainly you can still evade the system if you want to (just like currently), but it changes the tone of the thing, and I think at least a reasonable proportion of people who want multiple accounts would be happy to comply with this, and that the people playing against them would be happier for the openness.

I see where you're coming from, but think about this: Say most people are open about having multiple accounts. Now there are others who would prefer to play with a single account (Myself included. I still think multiple accounts is, at best, frivolous.). But now think of a third group of people who would continue to hide their multiple accounts so that they can use them to obviously cheat, i.e. play against themselves to inflate their own elo. I don't think people will stop doing that just because others are open with their mutiple accounts. I especially feel that way considering how long it took for the people in this thread to even admit they had them, even though they claim to have done so for "benevolent" reasons.

I was mainly presenting 1') as a strict improvement to 1). As 2) has pretty big technical issues I thought that one was a non-starter (disclaimer: I'd probably prefer 1) to 2) even if it were implementable, but I think that would be much closer).

In any case I'm not so surprised by people being slow to admit accounts in the current system (although some said that they had admitted them at various points, but not perhaps in a public declaration): they were against the printed rules. It will be greatly socially easier to admit if it's not against the rules, but encouraged by the community.

Now, there may be this third group of people at present, and there may be in the future. But I don't think that anyone who has admitted to multiple accounts would fall into the category of using them deliberately to inflate ELO! And I don't think we have any real evidence at this point that there exists anyone in this third group. My personal guess is that it has happened occasionally, but the perpetrators are not established highly ranked players. The game of Wesnoth has much more depth and interest than the game of cheating-the-ELO-system, and I don't think people who prefer to play the latter game are likely to stick around in the community for so long.

[Caphriel]

True, yes. But couldn't it be possible to associate the IP address of the user with his account name each time he logged on? For instance, say Insinuator logs on as 172.65.14.22 on Monday. The site logs that address. Now, say Insinuator tries to log on an hour later as Caphriel. Assuming Insinuator hasn't reset his Internet connection, his IP address should still be 172.65.14.22. The site would compare that with the last log-in from that address and block it. Someone could get around this, of course, but it would provide a level of security.

And, of course, there is no way to distinguish between two people on the same computer. That's a little obvious. Unless we ask Big Brother to step in...

No, actually, that won't work. You can't make the assumption that Insinuator hasn't reset his internet connection, because it's not always a matter of the end-user resetting it, but that's irrelevant. So, Insinuator could connect from 172.65.14.22, then disconnect. The site doesn't know 172.65.14.22 has been freed. Caphriel connects to the internet and is assigned 172.65.14.22. He tries to log in to the site. The site checks its logs, and decides that Caphriel is actually Insinuator, and blocks his log in attempt.

So not only would people be able to circumvent it, as you pointed out, but it could/would generate false positives. This is of course up to the ladder admins to decide, but given the problem domain, I think it would be preferable to have false negatives (failing to detect an alternate account) than false positives (incorrectly identifying a unique account as an alternate account.)

And, as you agree, there's no way to handle the multiple-users, single-connection issue. Attempting to implement a unique-connection based blocking algorithm would restrict ladder accounts to one-per-household. Again, I don't think that is particularly desirable behavior.

[anoel]

to this ip-discussion and other control ideas, let's remember how you could cheat without multiaccounting:

• change the wesnoth code
  change game settings (if the opponent doesnt notice)
  report wins vs inactives
  report wins twice
  try to use dc in your favor
  make deals with others (let someone win)
  maybe even hack the ladder database

all in all my point is: this community is based on trust. the central question is not how to get more control, but do we like to have several accounts per player or not. after this decision we emphasize all to accept and follow it - whatever it will be.

[The Black Sword]

I'm not sure why I keep posting here, I just get ignored anyway...

@SalsaRocoto, As I and I think others have mentioned, that effect is no different than what would happen when a strong non-ladder player joins the ladder for the first time. The fact that account is an alias makes absolutely no difference to the system.
As Scatha mentioned, the ELO system is set up to deal with this, the actual figures are interesting but I don't think it has much relevance to the current debate.

@Anoel, I totally agree.

[Insinuator]

So not only would people be able to circumvent it, as you pointed out, but it could/would generate false positives. This is of course up to the ladder admins to decide, but given the problem domain, I think it would be preferable to have false negatives (failing to detect an alternate account) than false positives (incorrectly identifying a unique account as an alternate account.)

Yeah, I was worried about the potential for false positives. That would really be unfair. Perhaps a geographical trace could be initiated additionally at the moment of log in. If you IDed a computer's location versus the same IP somewhere else, that should eliminate false positives. That's more difficult, and possibly illegal, though.

Another idea might be to use a cookie to tag each computer. The cookie is unique to each account name and checks to ensure that it is the only cookie of it's type on a specific computer. Of course, this would give people who use the same computer a bit of trouble and those who used public computers (or computers that refuse cookies) to play Wesnoth wouldn't work.

Hmmm. There must be some solution. Other games have means of preventing this, I'm sure.

EDIT: As a side note, has anyone else had the problem I posted above with the Ladder site? It is still happening...

[Caphriel]

Geographical trace is legal... and generally done through IP address. Because of the way IP addresses are assigned, you can usually narrow down someone's location to a relatively small region with a fair degree of certainty. However, that also means that another form of location identification would still result in similar false positives...

Using cookies is somewhat reasonable, but also somewhat insecure, and has problems as you pointed out.

The only effective solution I know of involves money. Charge a fee for an account and suddenly people don't want more than one. Additionally, once money is involved, you generally have a name and billing address. This is the solution commercial games tend to implement; those that care about controlling multiple accounts, anyway.

If you'll forgive the digression, speaking as someone who ran a competitive game server with a rating system, preventing duplicate accounts is generally more trouble than its worth. Automated detection is difficult to implement, and imperfect. Manual scanning is difficult and time consuming, and people will still slip through anyway. However, detecting people actively misusing alternate accounts in order to manipulate ratings is somewhat easier. That'd be things like players reporting games suspiciously close together over a long period of time, accounts that have only played against one other player and always lose... things like that.