After all these years ...................

Originally posted by wormwood
the odds of 'happening' to follow ONLY engine games over dozens of games, which is something the engine detection would pick on, are astronomical. it won't happen by chance before something like all the matter in the universe gets sucked inside black holes, and hawking radiated back out. which is a very. long. time. which makes it also a very bad excuse.

Who said anything about it being an excuse? I was just responding to the point made that following Weyerstrass' games might be dodgy. I don't think I'll end up following engine only games by using my DB, the top CC guys aren't slaves to their engines and I only use it for openings anyway. I would be surprised if I was still in the database after about eight moves.

Originally posted by Diophantus
Who said anything about it being an excuse?

profylactic move. if I hadn't explicitly stated it, someone would've asked. someone will probably still ask within a page or two, as usual when these things come up.

I wouldn't worry about using published analysis, rhp games etc. At some point you will be on your own in a game and if you are playing fairly it will show.

Originally posted by wormwood
profylactic move. if I hadn't explicitly stated it, someone would've asked. someone will probably still ask within a page or two, as usual when these things come up.

Ah, so someone will ask if they can get away with pleading diminished responsibility due to following the games of a known engine user?

Originally posted by Diophantus
Ah, so someone will ask if they can get away with pleading diminished responsibility due to following the games of a known engine user?

We have had something similar happen before- someone had built a database on the Traxler using engines and then was following it during his games.

I have a private cyborg database of 66.4 million engine vs engine games which I quickly constructed in the last 4 years.
This is not published anywhere online & if my moves all match an engines then this is the reason! 🙄

Correctly done the 3 match up system does not kick in till after
known theory has been passed.

June 2009 Squelch posted an excellent thread about the 3 match up
system and how it works.

How to detect engine cheats - a guide

Thread 114715

Originally posted by greenpawn34
Correctly done the 3 match up system does not kick in till after
known theory has been passed.

June 2009 Squelch posted an excellent thread about the 3 match up
system and how it works.

[b]How to detect engine cheats - a guide

Thread 114715[/b]

Now I see why someone on another site was moaning about this method being flawed. A statistician (a real one rather than an armchair example) would insist on random sampling for the very reason that Squelch gives for choosing games objectively. Non-random sampling leaves the door open for accusations of cherry picking the data! However, I can see an easy fix for that. All that needs to be done is to gather up all the games the suspect has played that meet the criteria and then take a random sample of those.

Originally posted by Diophantus
Now I see why someone on another site was moaning about this method being flawed. A statistician (a real one rather than an armchair example) would insist on random sampling for the very reason that Squelch gives for choosing games objectively. Non-random sampling leaves the door open for accusations of cherry picking the data! However, I can see an easy ...[text shortened]... the games the suspect has played that meet the criteria and then take a random sample of those.

Engine use tends to increase over time, rather like drug dependancy.
If you select games on a purely random basis you could be analysing games from months/years ago & before the player started cheating, so you get skewed results & a false negative.

Also, let's say you randomly select a load of games to analyse & find strong evidence that a player has used an engine in the end results.
How would you prove to a third party after the event that the games were selected on a random basis without them witnessing your random selection process?
They could easily say that you cherry-picked the games & ignored those with low match up.

This is why the simplest, most objective selection criteria is x amount of the most recently completed games vs high quality opposition, which all either have for instance 20+ non- www.chesslive.de database moves.

If you apply the same criteria to every suspect, then the accusation of subjective game selection or cherry-picking results is laughable.

Originally posted by Zygalski
I have a private cyborg database of 66.4 million engine vs engine games which I quickly constructed in the last 4 years.
This is not published anywhere online & if my moves all match an engines then this is the reason! 🙄

Humour aside; per the site rules, engine vs engine or human vs engine games can not be part a valid reference database. In practice, a few inclusions won't make much difference. But 66.4 million might...

Originally posted by Squelchbelch
Engine use tends to increase over time, rather like drug dependancy.
If you select games on a purely random basis you could be analysing games from months/years ago & before the player started cheating, so you get skewed results & a false negative.

Also, let's say you randomly select a load of games to analyse & find strong evidence that a player ha ...[text shortened]... spect, then the accusation of subjective game selection or cherry-picking results is laughable.

Actually it is not as laughable as you might think. If you objectively select games for every run of this procedure then it can be argued that it is no surprise you catch cheats because you pick games to fit your hypothesis. The argument would go you picked the games by running them through your procedure and discarding those that did not give you the result you wanted! How do you prove to a third party that you didn't do that? You can't, but it may not matter if the method works.

When I was suggesting you gather all the games that match the necessary criteria I meant all the games that match the criteria you use to select objectively. So you pick all the recent games that have 20+ moves out of database and are against suitable opposition. You then randomly select a sample from that. It can still be argued that you fixed the results but it is harder to argue that if you have some randomness in the selection procedure. On the other hand, your method is already arduous enough without adding more work for a volunteer. So, I can see why you might get complaints but suspect the effort involved in convincing one or two doubting Thomases is simply not worth it. In the end, there will be some you simply cannot convince, even if they are in the same room while you do your thing.

There is also the simple fact that your method is inherently repeatable. Anyone with a reasonable PC (even my £19 machine would do I suspect) and a modern engine should be able to repeat your results given the games you worked with.

Originally posted by Diophantus
A statistician (a real one rather than an armchair example) would insist on random sampling

no.

random sampling doesn't mean you automatically get 'good data', ie. data that represents accurately the phenomenom you want to study. so you prune out 'noise' & 'garbage' first. standard procedure for all kinds of signal processing. in the real world all data is imperfect. in order to get more accurate results, you clean the data prior to analysis. there are myriads of different ways to do that, depending on the problem.

Originally posted by wormwood
no.

random sampling doesn't mean you automatically get 'good data', ie. data that represents accurately the phenomenom you want to study. so you prune out 'noise' & 'garbage' first. standard procedure for all kinds of signal processing. in the real world all data is imperfect. in order to get more accurate results, you clean the data prior to analysis. there are myriads of different ways to do that, depending on the problem.

There is a difference between cleaning the data and selecting data to fit the theory being tested.

Analysing the 20 most recently completed games vs 2000+ RHP rateds which are all found to have 20+ non-database moves is less open to abuse than taking all games that fit the criteria & then randomly selecting some from the sample.
I'm at a loss as to what your reasoning is, since you can't prove randomness after the event & this approach does lay you open to accusations of cherry-picking.

Originally posted by Diophantus
Actually it is not as laughable as you might think. If you objectively select games for every run of this procedure then it can be argued that it is no surprise you catch cheats because you pick games to fit your hypothesis. The argument would go you picked the games by running them through your procedure and discarding those that did not give you the res ...[text shortened]... ct) and a modern engine should be able to repeat your results given the games you worked with.

In practice, the Game Mods here have tools that allow batch analysis and run a significantly large sample that this isn't a real issue.