August 2017 Anti-Abuse Report

August 2017 Anti-Abuse Report

monitor
monitor
Sep 12, 2017, 5:03 AM |
25

Chess.com’s goal is to provide an enjoyable chess experience for players and fans around the world. A large part of that effort is to protect the game from people who might try to ruin it for others with computer-assisted cheating or abuse.  

We can honestly say that we do all we can to meet this challenge, to protect the integrity of the game, and ensure that Chess.com members are as free as can be from this worry. The main challenge we face is that cheating and abuse are practically impossible to prevent ahead of time. Just like how the police cannot predict and stop crimes before they happen, we can only act after the cheating and abuse has occurred, which means it does affect some members.

We acknowledge that it remains an ongoing battle and one of the best ways we can handle it is to be open with our community!

We have been cautious with what we reveal about our systems because we do not want to give cheaters and abusers extra information on how to get better at it. The more we discuss our methods, the less efficient those methods will be in the long term. So, while there's a lot we cannot discuss, there remains intense interest in this topic and we therefore feel we should inform our members as much as possible to show the community the kind of time and effort we put into anti-abuse measures on a monthly basis.

Before I go into the numbers, I will speak a little to the science of chess cheat detection.

A frequent question I get is whether all cheating is detectable. The answer is, unfortunately, no. The most reliable algorithm in the world, under the greatest minds, could not detect if, on move one, you consulted an engine and let it calculate 1.e4 for you, and then proceeded to play the game yourself. A human could just as easily have played this first move.

While that may seem like a silly example, it's getting to the nature of the problem. Cheating is easier than detecting it, just as encryption is easier than decryption, or destroying a house is easier than making it.

So how do we approach this problem? The main proposition we work with is that engine play has statistically different patterns to human play, and with the correct analysis, over enough moves, these patterns can be distinguished.

This means that even if humans rose to the strength of engines tomorrow, we may still be able to determine that someone was cheating because the nature of the play would be different. This method also manages to catch most attempts to deceive detection by "smart" cheaters, such as those who choose second-best moves, or only cheat in spots.

The actual elements of the statistical side of our systems -- involving algorithms, modeling, data collection, background processes, etc -- will remain secret. However, our systems undergo constant scrutiny, review and improvement and have been vetted by Harvard statisticians, mathematicians and grandmasters. We continue to operate under the presumption of "innocent until proven guilty".

Feel free to leave me any questions or suggestions in the comments below.

Sincerely,

Gerard Le-Marechal

Cheat Detection, Data Analyst, and Statistics Research


 

Report for August, 2017

* Analyzed ~300k games total, or around 9 million total comparisons of Chess.com moves vs engine moves (~10k games per day)

** From which we banned 2,807 accounts for cheating.

*** Of which 2,497 were from live chess.

*** Of which 310 were from correspondence chess.

* Oversaw and deeply analzyed five major live events, Titled Tuesday, SCC match Grischuk vs Rapport, SCC match Nepomniatchi vs Aronian, SCC match Caruana vs Yifan, SCC match MVL vs Xiong

* Handled 75 appeals, in which we performed a detailed review of the cheat cases of closed players post-mortem

* Banned 683 IPs (IPs are banned of repeat offenders for both abuse and cheating)

* Closed 16,076 accounts for registration abuse, spam, abusive behaviour, cheating etc.

 Professional Relations Update:

We also help verify new titled players who join Chess.com.

* Confirmed 71 new titled player accounts (5 GMs, 3 WGMs, 17 IMs, 3 WIMs, 28 FMs, 4 WFMs, 4 CMs, 1 WCM, and 7 NMs)

* Updated seven titles as they moved up in the chess world: four GMs, three IMs (https://ratings.fide.com/title_applications.phtml?pb=49)

* Banned five titled players for cheating

* Denied the titled certifications of two accounts (impostors using fake credentials).

 

You can read about our site rules here:

* https://support.chess.com/customer/en/portal/articles/1444879-fair-play-on-chess-com-what-you-need-to-know?b_id=12321

Our official stance on fair play and cheat detection:

* https://www.chess.com/article/view/chess-com-fair-play-and-cheat-detection

 

Contacts if you have a problem:

* To contact customer support: https://support.chess.com/customer/portal/emails/new

* I'm available most days if you want to send me a private message: https://www.chess.com/member/monitor