Why don't these statistics disprove white's supposed larger (practical?) advantage in chess960?

Question

I've read that, at least theoretically, white has a larger advantage in 9LX as compared to chess. (See appendix.) This seems to be assumed in practice to the point that in some 9LX tournaments, such as the inaugural FIDE 960 world championship, they have players play both colours for any given starting position.

To even begin to talk about that white has more of a practical advantage, I think we should have some statistics that show there is a higher winning percentage change between white win and black win in 9LX as compared to standard. (Then afterwards we see if this increase is statistically significant or not.) But actually 'it's the reverse'! The winning percentage change is lower! See here (and hopefully here later on):

• in standard, white has 23% advantage against black: (39.2-32)/32=0.225, but

• in 960, white has only 14% advantage against black: (41.6-36.5)/36.5=0.13972602739

• (By advantage i mean percentage change between white win rate and black win rate)

In particular those are statistics based on engine games. My thought is that these statistics show how far theory and practice are here given that even engines, under time pressure, do not make the most of the white's 1st move theoretical advantage.

Question: What's going on? What's the basis for all this practical/theoretical advantage in 960 or at least for the both colours thing in 960 tournaments (perhaps it's not about white's advantage or perhaps they're just going on the theoretical advantage given lacking data on practical advantage or something) ?

Guess: Wait I think I just figured it out. Even if in 9LX white has a lower advantage (again, percentage change between white win rate and black win rate), say, in the above case of 14% vs 23%, white still has a higher win rate like 41.6% vs 39.2%. So, like, white's comparative advantage (lower percentage change) to black is decreased, but white's absolute advantage (of higher win rate) is increased...?

In this case I'd like to submit for consideration that comparative advantage be used as a counter-argument to absolute advantage in saying they must play both colours in 960 tournament or something, but I guess I'd save this part for another post and perhaps not on stackexchange.

---

Appendix on white's theoretical advantage (I actually didn't think I needed to cite because I thought this is why they do the both colours thing)

1. Starting positions in Chess960 where black is definitely worse

2. https://www.chess.com/article/view/whats-the-most-unbalanced-chess960-position

3. What are the best scoring openings in Chess960?

4. From here (Why isn't Chess960 the standard?):

Another problem with Chess960 is that some positions are more unbalanced than the traditional starting position. While some positions are also more equal, there are a few positions which start with more than twice the "normal" White advantage.

5. From When are we getting a World Chess960 Championship with classical time controls?

There are some positions where White has a statistical chance to win of more than 60%. That's 6-7% more than standard chess.

Answer 1

For the time being, without being certain one way or the other, I simply challenge the hypothesis:

You could look at eval tables and computer games (or human games, if a sufficient number are available separated by SP) to determine whether the White advantage is greater in Chess960, whether on average (mean, median) or how imbalanced SPs can get. Some limited evidence that the White advantage could be smaller in Chess960: http://www.computerchess.org.uk/ccrl/404FRC/opening_report_by_white_score.html (note the very high position of SP 518 in the ordered list).

• [EDIT: The Sesse evals for the 960 SPs vary from 0.00 to 0.57 and seem to confirm this, as SP 518 is evaluated at +0.22, ranked around 310/960 positions in terms of imbalance.]

As for why "they do the both colors thing" - well, as far as I know, they don't necessarily. There are plenty of Chess960 tournaments (including the actual Chess9LX tournament for example) where they don't swap colors on a given SP against the same opponent.

Finally, I don't believe that, when used, this rule is implemented primarily because of imbalanced starting positions. I believe in general the motivation is something like 'players A and B have just played from SP z, and we don't really understand SP z well enough to know how much of an advantage this really conferred the White player, therefore let them play again with colors reversed'. For example, even if - and it is a big if - we have engines whose evaluation we can trust as much as we do our opening knowledge of SP 518 (the standard chess opening position), it is still possible that humans would find it much easier to play from White than Black, whereas we know very precisely the extent of this inequity in SP 518 thanks to enormous experience of human play from SP 518.

Note also that the rule is more or less automatically implemented in match play in standard chess (the exception being Armageddon).

Answer 2

I suspect that we can't really use human (or computer) performance as a barometer of advantage in 9LX games. Standard chess has been studied for hundreds of years, has boatloads of theory, has plenty of heuristics like developing knights before rooks, etc. A lot of this doesn't apply in 9LX. So humans (and maybe engines less so) will consistently play less optimal moves. Imagine you and I race a Ferrari against a Porsche and the Porsche wins. Does that mean Porsche is better? We can't really say since you and I are not professional racers and really didn't achieve peak performance.

I'd guess this applies to neural nets to a lesser degree and to traditional engines even less so.