irwin VS lila

For Lichess, you can get some sort of idea from reading the code. If I remember correctly, there's some threshold for the site suspecting you of cheating based on a variety of simple metrics (accuracy, blurring, etc) and then it gets sent to machine learning tools to analyze (here is one of them, and here's another).

Lichess is open source. Here is their anti-cheating code on github: https://github.com/clarkerubber/irwin

Yeah Irwin is the old lichess model - here (https://github.com/clarkerubber/irwin) as well as the more maintained fork (https://github.com/lakinwecker/irwin/commits/master)

This is nonsense. They’re a free and open source nonprofit with no interest in commercial buyers. I’ve played on lichess for 3 years, over which time I’ve never encountered more than 1 cheater within 10 games (and normally much less than that). Because the website is open source, you can see the cheat detection they use. It’s not perfect — there is no perfect system — but it’s transparent and in my experience very good.

Isnt't lichess' cheat detection an ML based system?

What's this? It was given as an answer on a question about this on lichess https://github.com/clarkerubber/irwin

Once you’re finished, we’re going to set up a lichess bot account. Head over to https://lichess.org/ and create a new account.

Lichess - Online Chess Server. Scala, TypeScript

> the player who committed more blunders lost 86% of the time

In some sense this is almost tautological. While finding an exact definition for a chess blunder isn't straightforward, here is one example from the Lichess UI:

https://github.com/lichess-org/lila/blob/b527746b179cdde6438...

Basically, if you make a move which decreases your winning probability more than 14% over the best move, that's a blunder. But winning probability is a nonlinear function of stockfish centipawns. A drop in 100 centipawns when you're up 15 points isn't a blunder. When the game was equal, it is.

Point is, by the time you know it's a blunder you already know something about the outcome of that move, that it swung the winning probability by more than 14%. So the analysis is kind of just measuring some function of winning probability and saying that it is highly correlated with winning probability.

If you want to improve make it your goal to play the best chess you can, not increase an arbitrary number. Watch YouTube series like John Bartholomew's "Climb the Rating Ladder" for some general insight into what you might be doing wrong. Read Irving Chernev's "Logical Chess: Move By Move" to see the thinking process of high level players. Do lots of puzzles (I like lichess.org for puzzles). And always analyze your games. When you analyze make it your goal to find at least two things you could have improved.

The easiest way to play against Stockfish is perhaps on https://lichess.org/, but it's not the only chess engine that evaluates positions with a neural network.

Lichess.com

if you actually like chess, try lichess.org, the free and open-source, no ads ever, premium alternative