While racking my brains to figure out how a bot could be programmed to both use engine moves, while avoiding commonly used methods of cheat detection, such as T-3 moves or average error, or anything attributed directly to either accuracy or error, I closely examined the differences between the way humans choose moves and the way engines choose moves.

Humans will use zugzwang in positions where they are capable of calculating it, particularly in mating net scenarios. However, in positions where they cannot calculate whether or not which of the options, they leave their opponents are certainly winning or losing, they don't bother trying, they simply look for the move that seems to give them the best chance to win if they see it, or the best move to not lose, if winning seems to be not possible.

An engine on the other hand, can calculate with relative certainty, whether or not a move leaves their opponent any moves that give a winning advantage, or that allow for the possibility of a draw. When a human plays an engine at full strength, the engine simply makes all of its choices in the most forcing lines, which limit the options that its opponent has. This lowers the probability of their opponent finding their only playable option, if they have been reduced to only one move, for example, they can play to win or draw. Bots in particular can take advantage of this, if they can pick a move from a list of moves that gives the best combination of the fewest playable options (closest to zugzwang they can get their opponent), while simultaneously avoiding best move match up rates. This is easier for the bot at fast time controls particularly, due to the fact that even masters tend to play moves with an average error between .5 and 1 pawn in fast games.

Here is a game between a banned cheater and a FM in 3 min blitz.

fartingfrench vs. F1-Herakles | Analysis - Chess.com

I pulled off a random banned cheater from a FM's archive, who only played 3 min games. I also picked a random game where the stats looked lackluster.

Here are those stats:

It could be argued that the t1 moves were a bit high for both players in a 3-minute game, but these totals are unfiltered and the forcing nature of them isn't accounted for in these totals. However, looking strictly at the stats in comparison to one another, I suspect most PGN Spy users and T-3 proponents would conclude that there was no cheating in this game by either player. However, I have evidence that suggests otherwise.

Imagine if I told you that the first 22 moves of this game played by white didn't give black a single move that it could play to gain equality or a winning position? The game left Chess.com's opening book after 7 moves by white and left Chessbase's database after white's 12th move. During the time black played its first 11 database moves, which includes moves GMs used in their games, it gave white winning chances with 63.6% of its moves and drawing chances 5.5% of the time. How is it that white's moves came from the same games for the first 11 moves, but didn't give black a single move to gain equality or an advantage? Surely masters playing with either color would choose moves that gave their opponent playable options during the database portion of the game, right? Perhaps it could be argued that the top 10 super GMs might be able to play accurately enough to prevent their opponents from having a chance to gain equality and an advantage in the first 22 moves of game, but someone without a title, while playing a FM? I can't say definitively that this game in particular got @fartingfrench banned. However, I wouldn't at all be surprised that his opponent did report him after the game on suspicion of cheating. To shut down a FM for the first 22 moves and then crumble, getting checkmated in the last 6 moves is rather suspicious for me personally. This is exactly what a friend of mine used to complain about in 3 min games. They would play flawlessly, even if the post-game stats or PGN Spy didn't or couldn't prove it, then they would crumble once time got low and the position became simple enough that the human could pre-move, and the cheater had no time for the bot to calculate the moves it needed to play to win.

This is how I am isolating the engine using zugzwang. I am using Arena with Stockfish 4.5 DD. I could use a newer version, but for me that isn't as important as what I am about to show you. I use a multi-PV analysis of the 15 best moves in each position. I look at each position exactly 20 moves deep. PGN Spy will not look exactly the same depth in each position. I have tried repeatedly using various settings, and it simply will not. I find this gives inconsistent results. In some positions where clarity is really needed, it won't look deeply enough and in other positions, it will waste time analyzing them far deeper than necessary. Here is an example of what that analysis looks like.

Here is where this gets a bit technical. So, these are the stats from black's 24th move, but the moves from white's move 24 are what left black with these options. So, white's 24th move left black two options that give a winning advantage, 5 that give the possibility of a draw and 8 moves that are losing. So, I am about to post the analysis results put into a spread sheet and it is important to understand that the first move stats simply don't count, as neither white nor black are responsible for the nature of chess and what any of those moves leave black. White has to play something. However, white's first move choice does affect what options black has and, so the stats start there. Also, once completed, the numbers that show up under white's moves, are actually black's stats and vice versa.

*note- the options white allowed show up under black moves and vice versa, so in essence white's stats are actually on the right and black's stats are on the left
**null=checkmate, thus no stats

So, this concludes that even though white lost the game at the end, white only gave black 3.2% of the possible moves they could make, that give a winning advantage. By contrast, black, a FM, gave white 51.11% of the moves it could play, as a winning option. In my opinion, this game is highly indicative of cheating all on its own and doesn't need supporting evidence. If this game was between two super GMs at this time control, I would want more evidence. If this game was between two GMs at a rapid time control, I would want more evidence, especially if it wasn't a longer rapid format.

For more evidence, more games could be looked at and players compared to known clean play from OTB for similar time controls or known clean play in online events. I am under the impression that Chess.com monitors and uses some of its titled players on its staff and CDT for the purpose of obtaining clean online benchmarks for various levels. Obviously, white tends to be rather forcing with its moves towards black. However, I am convinced that if you looked at human play at various levels, it would be quickly concluded that humans give far more winning, or drawing options to their opponents, especially in the late opening/database phase and the middle game, than an engine. This would become quickly apparent by both average options given, both per move and per game, and the percentages overall.

@ElCoachRene, feel free to use this to police your own games, but please don't share this publicly. Cheaters absolutely must not understand this, or they will particularly be able to make use of it at longer time controls. Bot programmers I believe already understand this also, but I believe they may try to find ways to refine how the engine can make use of it in their programs. I believe they would tend to give a few more playable options, banking on the probability of mistakes, once it assesses the average error and number of options its human opponent gives. it will then have a better idea what it can see and then like Mikhail Tal, uncork complex combinations that full strength Stockfish finds flaws in, that the bot knows the human is very likely to not solve.

@Monitor, I will never send another report in for a player, unless they show an obvious pattern of using zugzwang at inhuman levels and I only plan on sharing this with a very small group of people, whom I will both respectfully ask and demand their silence.

Non-Existent

Super Admins