Cheating detection absent in lichess

@for_cryingout_loud said in #30:

pls site source and funny enough you have been proven wrong in this thread
#21 i meantion the fact of how a person with 6k games found none but a person with 3k gets them every 200 proving that it is to small of a sample size but if your sample size for your data is a fraction of a percent and you want to tell me it means something it does not. Never mind the problem of all the data being from 1 person or the fact we still need to analyses the data

Cite.... how statistics works? Well, here's a website you can plug the numbers into. It'll show how it arrives at the result at the bottom. Calculating this sort of stuff by hand every time is... not fun, especially since I don't do it often. https://www.statskingdom.com/50_ci_sample_size.html

Basically, if you perfectly randomly select from a population (the entire group—in this case, all ~100000000 games), you will expect whatever feature of the games you're looking at to be present in a similar proportion in your sample as it is in the whole population. This is because, the more you select (the larger your sample size), the less likely it is that you've just randomly happened to select outliers.

For simplicity's sake, assume we have a bag with one hundred million marbles. Each marble is either red or yellow. How can you figure out how many there are of each colour? What I think you mean by "analyzing the data" is checking every single marble. This is not feasible (on Lichess, I know you can download data—I don't know if it includes whether a player was banned for cheating. As I'm talking more generically about how statistics works in general, and I don't particularly care about the result, we'll say it's not. But even if I did, I would choose a smaller sample simply to save time, because processing data from that many games could be tricky). Since it is not feasible, we need to figure out another way of figuring out how many red and yellow marbles there are.

If we grab one marble, it'll be red or yellow, but that will only tell us that there are more than zero marbles of that colour. Let's assume they're split 50-50 by colour, as that requires the largest sample size to confirm (play around with the Proportion field on that site I linked to see how that changes). If we grab ten and we get six red and four yellow, that's not particularly unlikely—99% of the time, a sample would have between 1 and 9 red marbles. If we grab a hundred and get sixty red and forty yellow, it's not too unlikely, but it's less likely—99% of the time, a sample would have between 38 and 62 red marbles. If we grab a thousand and get six hundred red and four hundred yellow, that's unlikely if they're actually split 50-50, as, 99% of the time, a sample of a thousand would have between 460 and 540 marbles.

Note how the range of expected values goes down as our sample size increases. Initially, we could expect a range covering 80% of the marbles. Multiplied our sample size by ten and we had a range of just 24%. By ten again and, 99% of the time, we'd expect to draw a number of red marbles across a range covering just 8% of the sample size.

By flipping some stuff around with the underlying equations, you can calculate the most likely proportion of marbles to have resulted in you grabbing however many red ones you got. If you only sampled ten marbles, the range of possible proportions would be quite large. If you sampled a hundred, it would be a fair bit smaller. If you sampled a thousand, it'd be even smaller. The actual size of the population doesn't matter too much—on that same site, you can also modify the size of the population and you'll see that it has little effect. From one hundred million to infinitely many games, you'd only need three more samples to be just as certain of your results.

You claim that I was already proven wrong in this thread, based on someone saying that they get points refunded from about 0.5% of their games. However, you're assuming that the "6k games with no cheaters" claim is correct. They might not have ever had points refunded for cheating, or maybe they did and forgot or didn't notice, or maybe the cheaters they played were caught more than 40 games or 3 days after playing them. If you actually look through their 6000 games, I'm sure you'll find a similar proportion of games against cheaters as in your own games.

12 year-olds love cheating on game sites ... it's very chimpadelic ... if I was 12 I would probably do it too ...

Can we delete (move to a different location) all of the lichess database and start a new one with an incredibly powerful cheat detection tool that will also remove all of the 30+ 2800s from the website please?

How ironic.

@Morobiondo said in #1:

Over 6K games played and no chetating detected. In chess.com (free) they detect and find this kind of abuses...

Maybe because there are more cheaters on Chess.com

@Morobiondo said in #1:

Over 6K games played and no chetating detected. In chess.com (free) they detect and find this kind of abuses...

Ask Kramnik!

@dokoko1 said in #20:

I am regularly refunded rating points for rapid games lost to people who violated the lichess rules (by cheating I guess). My rough estimation is one game in 200.

Did you subscribe a pay plan? Or made some donations? Just curious...

Just 10 day ago:

We have detected that one or more of your recent opponents has violated our Fair Play Policy. As compensation for potentially unfair rating losses, we adjusted your following ratings:

Rapid: 1080 + 9 => 1089

In my opinion this is at least somerhing a kind of control on cheating compared to the ZERO of lichess

@Morobiondo said in #38:

In my opinion this is at least a kind of control on cheating compared to the ZERO of lichess

Did you read and understand https://lichess.org/forum/general-chess-discussion/cheating-detection-absent-in-lichess?page=3#26 ?

@Morobiondo said in #38:

In my opinion this is at least somerhing a kind of control on cheating compared to the ZERO of lichess

No one is cheating to play at your rating. If they were cheating they would be climbing rating quickly - most cheaters will start at 1500 and go up, not down. It is quite possible that you have never played a cheater.

If you think high accuracy indicates cheating, look to see if you make any mistakes. If you blunder, it is easy for the opponent to find strong moves, which makes the accuracy look high when you lose. But the mistakes are yours to own, don't look for excuses.