Breaking Down 100,000 Closures a Month
Our 30-person Fair Play team at Chess.com is now closing over 100,000 accounts a month. While this number represents an exceedingly small percentage of the total accounts on our site, the sheer volume of closures has risen as more people have created accounts on Chess.com.
(Note that 0.3% is not 30 out of every 100 but is instead 3 out of every 1000!)
One of our primary challenges as a team is understanding the evolving trends and patterns in account behavior so that we can more quickly identify and close accounts that are cheating. You can analyze closed accounts in a myriad of ways, but one factor stands out more than many others: account age. Nearly 40% of accounts that are closed are no more than two weeks old! In short, new accounts make up a plurality of closures on site.
How many games are accounts playing on site before they’re being closed?
In the past month, the median number of games played by an account before being closed was 52 games. This actually reflects a 17% drop since December 2024, when the median games played for closed accounts was 63. CCC
This suggests that our screening and autoban conditions are catching cheaters faster than ever, but they could potentially be improved even further. At the same time, there may be real limits to how low this number could go, as it’s also reflective of how soon a player may become a bad actor. We can close accounts on less than 10 games if the statistics warrant it, so most closed accounts aren’t the product of cheating from the first game on the platform.
We can close accounts on less than 10 games if the statistics warrant it, so most closed accounts aren’t the product of cheating from the first game on the platform.
What can we do about it?
One of the ways we’re tackling this phenomenon is by screening players more frequently early in their journey on the platform. This means we’re generating reports for players that essentially serve as checkpoints as they play more games. It is entirely reasonable (and expected) that it may take time for someone to establish their rating, so these checkpoints go beyond games played and rating earned, but also include strength of play and over 100 other gameplay factors.
While our team diligently reviews and generates thousands of reports on a weekly basis, it is an impossible (and undesirable) task to review the sheer volume of games (~20 million played per day on site!) manually. We rely on carefully calibrated algorithms that enable us to operate at scale and help us determine what extraordinary conditions merit a closure.
For obvious reasons, we will not elaborate on those conditions here, but we can say that 85% of accounts are being closed due to autoban conditions that identify inhuman strength or playing behavior. This enables the team to focus our efforts on more complex cases that don't get autobanned and require further analysis.
What else are we considering?
Players who cheat sour the experience for players who are not cheating. That is well understood. Given that new accounts make up a plurality of accounts closed for cheating, we’ve considered exploring additions to our matchmaking system in pool play, where players need to establish themselves with more clean games on site against other newer accounts before playing more established accounts. This idea may have some cons (newer accounts, for instance, represent a smaller percentage of overall accounts on platform, making it more challenging to get a game), but the pros seem pretty obvious. What do you think?
IM Kassa Korley
Senior Fair Play & Communications Advisor
