Very interesting read. I don't play video games, but I do play online chess, and in online chess, there is a huge epidemic of cheating. Many cheaters are banned by chess.com. Some of these bans go unnoticed, but in other cases, the cheaters passionately insist that they were not cheating. And I don't believe them. In a minuscule number of cases, chess.com has been known to reverse a ban. But chess.com does not provid…
I had the interesting experience of being banned from a Call of Duty 4 server, back when the franchise still had servers. It happened like this: we were playing the game mode Sabotage, and it went into overtime. When this happens, the game shows the exact location of every player on the map to every other player and prevents respawns until there's nobody left on one team, at which point the other wins. In CoD you can…
1. There is no human player at all, just a bot playing the game (maybe being paid for by a user to boost their rank). This is likely detected by all the usual anti-bot heuristics that many web services have. Another option might be looking for statistical outliers on how highly player's moves correlate to open engines like Stockfish (this was the cause of a big cheating scandal in pro chess last year, if I remember right)
2. There is a human player, but they're just feeding the moves into an engine like stock fish and copying them out. Again, this is probably based on statistical correlations.
Here's the thing with any anti-cheat, the standard for a scientific paper is based on a P value indicating the likelihood of something being chance is about 5%. This is obviously way too high a threshold for anti-cheat, it would make 1 in 20 of your bans false bans. But the logic of an "acceptable" heuristic about lucky shots, or headshot rate, or blink stalker micro saving low hp units, or stockfish-correlated chess moves is part of basically all anti-cheat systems.
I'd guess they tune their thresholds to be something more like 1 in 1000, but after a point the way you reduce false bans for these things is to ban less, which given the high rate of actual cheating, is not desirable to the game companies. So if going from 1 in 1000 to 1 in 10000 requires halving the amount of bans you dish out, game companies are just not going to do that.
So then some CS drone has to answer the ticket about why the user was banned. They know that 999 of every 1000 tickets are lying, so they just automatically close it with no recourse. It's not worth the company's resources to make a recourse process. For PR purposes it's better to just pretend that isn't the case, and say there's no false positives. We've seen the PR reaction machine initially respond the same way when the "is something hooking the game process" checks detect all the Teamspeak overlay users and bans them all, until the sheer volume of affected people cause them to relent. So it's hard to believe that when they have statistical modelling based bans affecting much smaller numbers of people, they don't just steamroll them.
Heck, it's not anti-cheat, but I've had a copy of Red Alert 3 basically stolen by EA as they claimed my CD key was pirated (it wasn't, I bought it on Steam directly). Of course CS claimed infallibility, but it's made me be a lot more suspicious of other cases where CS claims infallibility.
But these games have millions of players, so 1 in 1000 is... quite a lot of people actually.