When will we see HOUDINI in official tournaments?

[hgm]

I can just add code to play randomly picked good move when the new game starts. This is done just for the first move which should be enough to fool the similarity test.

Not at all. Because the moves played in the test from the positions taken from your tourney games would show no similarity at all to the moves you played from those positions in the tourney game. So you would be found out as a cheater, having provided the organizers with an exeutable that behaves differently in the test as during the games..

[LudiBuda]

Similarity test isn't doing anything you are talking about.
In the unlikely case CSVN decides to do what you are proposing, you can always say that different moves played from your tournament game are result of the MP search or hash table inconsistencies.

[hgm]

1) You can safely bet your life that the CSVN will do what I am proposing.
2) You can say whatever you want, just like they can ban whomever they want...

[Rebel]

Thanks.
The Similarity test could be trustworthy for commercial engines.
But it is no problem for a special CSVN-tuned engine to recognize beeing tested…
Isn’t it, Ed?
So the CSVN rule seems to be doubtful.

You will be surprised how hard it is to remove the playing style of a program without considerable loss of playing strength. Take an open source and try to get it below 60%, then we talk again.

[EmilV]

I don‘t need to keep playing strength in analysis position mode. Different next-best moves could be randomly chosen for example.
It is enough to play at full strength only whole games (or maybe better the lines without the first move).

[hgm]

And you will be found out immediately, because the program played different moves in its tourney games than when they later test it on the tourney positions...

[EmilV]

1. Immediately?
The different behavior in the same position is usual for MP engines. If the trick would be well done, after several tries you would get the same move as in the game. And there is not a simple way to repeat the whole game unrecognized by engine.

2. Even if so I don’t see any CSVN rule the engine has to play equal moves in games and tests.

[hgm]

What you suggest doesn't work at all, because the test is quantitative. It is the fraction of the moves that is the same which will be counted. That it is not 100% to begin with is immaterial. That after 10 tries you might get the same move as in the game doesn't help to get 70% of the moves the same.

Any attempt to lower the similarity with the engine you cloned by randomizing moves during the test will also lower the similarity with the engine that played in the tourney. The randomization by SMP effects will not nearly be enough to hide the similarity. There is a huge Elo gap between most SMP engines and a random mover...

[LudiBuda]

All fine and well until you get to the remote engines issue.
They can use one version to cheat on similarity test and another for the tournament and the tournament games check.

[hgm]

Not sure what you have in mind, or who you mean when you say 'they'. If someone plays with a remote machine, and the moves in his tourney games are 90% similar to the moves Houdini would play in the tournament-game positions, it seems to me he is caught...