Would it be accurate to expect a random legal position to tend to evaluate to a white advantage? How does the distribution of evaluations look like?
On those assumptions, it would probably evaluate to the benefit of the side the engine is playing: many engines have a contempt-factor to bias their own side.– user30536Nov 11, 2022 at 14:13
1The side to move would have the advantage on average. Is side to move part of randomization?– Michael WestNov 11, 2022 at 14:40
@MichaelWest Yes– 2080Nov 11, 2022 at 17:12
Given a legal position, the same position with colors swapped (and rows mirrored) is legal, apart from a very small number of exceptions like the position after
1.a3. So the average evaluation must be very close to 0.
1Actually, small correction: the position after
1.e4is legal even with colors swapped because black can lose a tempo with the queen or bishop:
1.Nf3 e5 2. Ng1 Be7 3. Nf3 Bd6 4. Ng1 Bf8. An example of a legal position that becomes illegal after swapping colors is the one after
1. a3, since one cannot lose tempi moving just the rook and knight. Nov 11, 2022 at 14:54