Need feedback from turn-based players to counter ELO boosting

[sourisdudesert]

Recently, a player did the following to boost his ELO at a game :

• Starting 200-300 games in turn-based.
• Play fast when he was loosing, play slow when he was winning.
• He manages to have a 30% win ratio for the first 150 games to finish (those with a bad position).
• Finally, he played the final moves of his last 150 games with a winning position, with 150 wins.
• Thanks to this impressive series of win, he gets over 2000 ELO and first place (which is unrealistic for this high-luck game).

(there is no need to give names here, the player ELO has of course been re-adjusted)

The best idea I have now to fix this is the following : limit to 30 the number of games in progress with ELO "on" for a given game (with 30 games, this technique cannot bring you a significant boost).

(Another idea would be to forbid you to play on a turn based table while there is another table with the same game and less time to play, but I really dislike this).

If you have any idea of a better solution to fix the problem, feel free to share it.

Note that the best solution would have been to reorder the results afterwards and to recompute all the ELOs based on the game starting date, in this case we should also recompute all the ELOs of their opponents, and so on, and finally we would have a situation when everyone's ELO would be very unstable (even with no game played) which is not acceptable.

[Imay]

Hi,

I can't think of an ideal solution, but as you said, a limitation on the simultaneous numbers of game would be some kind of it.
but i see a problem.
If I register to 4 tournament round robin? what would happen then when the fourth would start ?

[sourisdudesert]

If I register to 4 tournament round robin? what would happen then when the fourth would start ?

This would not concern tournament games. Tournament games are less problematic as the time constraint is way more strict so you have less latitude to stop playing.

Note that the limitation is per game, so you can play 30 Can't Stop games in parallel with 30 TTA games for ex.

[demiurgsage]

1. My idea is that ELO changes should be applied to the moment of start of the game, not a moment of finish of the game. Also turn-based games with ulimited time to play shouldn't be counted in ELO. In Elo games tournament timer should be applied. So all Elo games will be played in month maximum. You can further improve this system adding smothing to ELO, increasing influence of non-recent games.
Players can try to manipulate this system too, but their star hour will not last more than two weeks.

2. But maybe you shouldn't do anything about it. Just forbid this kind of behaviour and write a script that will check top players.

[sourisdudesert]

My idea is that ELO changes should be applied to the moment of start of the game

Yes, this would be the best option but it's not acceptable (see my "Note" in first post).

2. But maybe you shouldn't do anything about it. Just forbid this kind of behaviour and write a script that will check top players.

This is what we do know, but it costs time, which is the smallest resource I have :/

[demiurgsage]

My idea is that ELO changes should be applied to the moment of start of the game

Yes, this would be the best option but it's not acceptable (see my "Note" in first post).

This already was realized on one site with addition of smothing system. And actually Elo changes due to recalculation of old games were not more than 5 points from average value for a player with 50 games that I think is acceptable. Also recalculation was only once per day for all players. So your games that were finished today will affect your rating only after recalculation.

[demiurgsage]

2. But maybe you shouldn't do anything about it. Just forbid this kind of behaviour and write a script that will check top players.

This is what we do know, but it costs time, which is the smallest resource I have :/

This is not so urgent. Just add news that players who will start or continue doing something like that will be banned.

[demiurgsage]

I thought a bit more on your question, it looks like you found all possbile solutions. At least if you want maintain Elo for turn-based games.

[Jest Phulin]

Could you please explain your note? I'm not understanding why calculating ELO change based on starting ELO instead of ending ELO would require recalculating all opponents' ELOs also.

While I can see that a player may increase their skill during a game by playing several simultaneous games, I don't see this as a large enough change to affect everyone else's rating.

I would suggest just using the starting ELO of each player to calculate ELO change. Unless, of course, that figure is difficult to obtain as only the current ELO is listed in the database.

However, if no other solution is proposed, limiting the number of simultaneous ranked games is a good solution. The purpose of this site is to have fun playing games. There should be no difference in fun from playing games 1-50 at once, or 1-10, followed by 11-20, followed by 21-30, etc. If people think that there is a difference in fun, they can turn ELO off and play as many as they want.

[sprockitz]

I think there are a few things that can be done. Limiting number of Elo on games will certainly work to some degree, but I still believe you can get a pretty high Elo even with only 30 games at a time. Say you can fairly confidently get to 1600 in a high chance game, just by normal luck fluctuations. So after losing a bunch of turn-based games and starting new turn based games until you have 30 in the bank ready to win, you can work yourself back to 1600 playing real time games then cash in those 30 turn-based wins. Assuming you are picking opponent's to try and get 1600+ opponents you could win ~300 points and end up about 1900.

So...in my opinion, in addition to a games limit, there should be a K-factor adjustment. Right now we have two K-factors based on games played (1-30 and 31+ games). For higher luck games, in particular, this K-factor should go down so that the ratings will stabilize.

For example:
1-30, K = 30
31-60, K = 20
61-100, K = 15
101-150, K = 10
151-200, K = 7
201+, K = 5