So called even positions

[CRoberson]

I know that this will make it harder on those trying to create even positions, but so.

We have seen several discussions about even positions. Some of them turnout to be far from even. Some suggest that using an engine
for deep search is not right. Some have stated that some clearly uneven positions are even and the engine in question doesn't understand
the position well enough. All that leads to how to prove it is an even position. Also, what does even position mean?

I suggest take a position and a program and have that program play itself for 200 games or more. If the score is close then its an even
position. Maybe, it should be 50 games with 4 different engines playing themselves which equates to 200 games.

There is another problem. Some positions are even only under long TC conditions. So, testing under blitz or bullet conditions will
make them look uneven. You may think that is ok, because you end up with only even positions. No, not so. There are positions that
look even at blitz and bullet, but long TC's reveal them to be uneven.

Seems a good process would be a few hundred games of blitz to filter out some and then a few hundred games of G/120 to filter out
the rest. There you go that is the process. Now, make a list of guaranteed even positions. It is easy.

[bob]

I know that this will make it harder on those trying to create even positions, but so.

We have seen several discussions about even positions. Some of them turnout to be far from even. Some suggest that using an engine
for deep search is not right. Some have stated that some clearly uneven positions are even and the engine in question doesn't understand
the position well enough. All that leads to how to prove it is an even position. Also, what does even position mean?

I suggest take a position and a program and have that program play itself for 200 games or more. If the score is close then its an even
position. Maybe, it should be 50 games with 4 different engines playing themselves which equates to 200 games.

There is another problem. Some positions are even only under long TC conditions. So, testing under blitz or bullet conditions will
make them look uneven. You may think that is ok, because you end up with only even positions. No, not so. There are positions that
look even at blitz and bullet, but long TC's reveal them to be uneven.

Seems a good process would be a few hundred games of blitz to filter out some and then a few hundred games of G/120 to filter out
the rest. There you go that is the process. Now, make a list of guaranteed even positions. It is easy.

I don't think that's safe. Ideally you would take the same position, alternate colors and play a group of opponents. The see if the black/white win advantage is outside the norm. If not, it could be considered "balanced".