4NCL Online

[Roger de Coverly]

If a 1900 players make 4 blunders every 20 moves, and then we have has a string of games without a blunder, you can be as certain as can be of engine assistance (or consultation with better players).

I hope you don't seriously believe that, particularly in the context of over the board chess where players are under constant observation. Or for that matter, historical games before engines got to any strength. You can also draw the conclusion that the rating of 1900 is incorrect.

[Li Wu]

So carlsen has a blunder rate of ~0.5% over a huge sample (blunder defined as -100 centipawns). If he makes 0 blunders in next 100 games (he won't) I will brand him a cheat.

[Li Wu]

Actually I doubt sites take actual ratings into account (but it can)- as most of the players do not have real life ratings.

A new, unknown 2500 player signing up does not get snap banned- so their actions must be taken due to truly extraordinary performances and/or other data that isn't revealed to the public.

[John McKenna]

"The fact remains it's possible to brand people as cheats with a lower false positive rate than the criminal justice system based on moves alone."

Hi Li, you may not remember me but we played once - at Southend, Easter 2003.

You won that one easily, but I''d like to challenge your statement, quoted above, because in light of the fact that there is hard evidence that the "false positive rate of the criminal justice system" borders on the criminal and is consequently a bad yardstick with which to compare, in any case.

Best ..
John

[Li Wu]

Southend congress 2003! Wish I could play more these days...

Yep there's a lot of wrong in the world- not sure how to solve all of them. I personally think this yardstick is OK though, as the consequence of being wrongly banned from online chess cannot compare with being wrongfully convicted. As I posted before- people "cheat" for many reasons, and we shouldn't judge all cases the same- in particular (at least IMO), online cheating is not as bad as otb cheating as otb is for sure premeditated and never opportunistic (given some context).

One thing I wish the chess sites would share (and I think they can without giving away much) is some idea of their threshold levels for false positives, and systems (if any) that are in place to protect innocent players.

[John McKenna]

Thanks Li,

A criminal case in point is -

The Great Post Office Trial

https://www.bbc.co.uk/programmes/m000jfyv

On BBC Radio4 FM right now.

[JustinHorton]

If a 1900 players make 4 blunders every 20 moves, and then we have has a string of games without a blunder, you can be as certain as can be of engine assistance (or consultation with better players).

I'm not entirely happy with "string" here - you really need to be able to put a value on it - and I'd also like to add that it's possible for one's play to improve dramatically, both in ths sense of a large improvement and that of it happening overnight. Obviously there's some kind of limit to that, but beware of assuming guilt when what you actually have is an outlier, yes?

[JustinHorton]

To give an example of the sort of thing I mean: since the end of March I've been playing rapid chess on lichess, and I've played a lot of games. Prior to that, I hadn't played at all since September, and I don't think I'd played internet rapid chess in twenty years.

After a while, I noticed that I was winning almost exactly two games for every one I lost. (Although there's a few 4NCL and other games, nearly all of them are created and played using the same parameters.) This was quite useful as a target - I'd pull ahead of the 2:1 ratio a bit, then fall a littie behind and so on, and as it happens I got to 100 losses just before I got to 200 wins (I was on 198).

Since then, I've won 120 games and lost 30, improving my previous 2:1 ratio to exactly 4:1.

Why? Who knows. Maybe I've just got far more used to the medium, maybe 300 games was what it took to get my opening repertoire to a different level. I don't know.

But I do know that it's startled me, and that it's done so all the more in the knowledge that somebody might find it suspicious

[Li Wu]

Yup- should be aware of outliers.

I've actually done some amateur analysis recently using this excellent program- https://github.com/MGleason1/PGN-Spy/releases

I invite you guys to do some analysis yourself. These sites use much more than just moves for cheat detection if you snoop around a bit more, and likely use tiered systems for various buckets of player groups (new players, high blunder rate players, verified titled players- for chesscom).

For your win/loss ratio- this surely also depends a lot on their matchmaking?

Also for lichess itself it's well documented that ratings are not randomly distributed (shout out chesspit podcast where I first heard of this)- as lots of players quit when they reach xx00 rating (1800, 1900 etc with 2000 being the most popular one)- so a 2001 player might be "over rated" as its full of players who ran above expectation and quit to not go back down again. Dunno how big this effect is regards to your personal winrate.

[JustinHorton]

For your win/loss ratio- this surely also depends a lot on their matchmaking?

Sure, but I've no reason to think it's been particularly different the 4:1 period than in the 2:1 period. Not that I've checked, but it's the sort of thing I might expect to notice.

[Li Wu]

I would say that you've gotten used to the interface- and that there is more variance with rapid chess. That is, most of your opponent's have ratings that are less representative of playing strength due to lower samples themselves. But who knows- prob some other effects as well.

[Joseph Conlon]

Justin:

Has your rating gone up? I don't know what your seek parameters are, but if you are (say) searching for players within a rating of 200 of yourself, then if your rating improves your relative win ratio should as well, because you are further along the exponential tail of the distribution, and so (assuming they draw opponents randomly) you will face a higher ratio of lower rated opponents, and so expect to win a higher fraction of games.

And if your rating hasn't gone up, then something odd is happening if your win ratio doubles and your rating doesn't improve!

As one related normalisation: my rapid rating on lichess has mostly been in the 2200 - 2250 range, and on a sample of 70 games my win:loss ration is 4.5:1 (using their default quick pairing).

[Kevin Thurlow]

"I wonder if we can have something similar to expert witnesses- where the secret detection methods aren't revealed, but evidence can still be used"

Er, no. Expert witnesses can expect a good grilling from a slippery barrister on minute detail of analysis.

[JustinHorton]

Justin:

Has your rating gone up?

I'm playing casual (i.e. unrated) games

[NickFaulks]

I personally think this yardstick is OK though, as the consequence of being wrongly banned from online chess cannot compare with being wrongfully convicted.

That is not encouraging to read. Someone who made their living as a chess coach and now can't do that any more will think the consequence is serious.