#9672
"your thinking is that of a chess player, not a game theorist"
++ Game-theoretically there are only 3 possibilities:
A) Chess is a draw
B) Chess is a win for white
C) Chess is a win for black
There is a lot of evidence for A)
1) Expert opinions of world champions Steinitz, Lasker, Capablanca, Fischer, Kramnik...
2) A high draw rate of AlphaZero autoplay; even higher with more time per move, this even persists if stalemate is considered a win
3) A high draw rate in ICCF WC correspondence play, even higher over the years, despite the fact that 7-men endgame table base win claims are allowed that exceed 50 moves without pawn move or capture
4) A high draw rate in TCEC engine competition, despite the fact that slightly imbalanced openings are imposed to avoid all draws.
There is no evidence for B)
There are a few minority opinions of Rauser (1 e4 wins) and Berliner (1 d4 wins).
There exists long wins for white or black in positions with 8 men, but no evidence that these can arrive from the initial position by a reasonable game, i.e. a game with > 50% accuracy.
There is no evidence for C)
There exist Zugzwang positions with 8 men won for black, but no evidence that these can arrive from the initial position by a reasonable game, i.e. a game with > 50% accuracy.
"A sample of a few thousand games between imperfect players could never be reliable."
++ My calculation shows that 99% of ICCF WC draws are ideal games with optimal moves i.e. perfect play. My calculation starts from the hypothesis that A) is true, but there is no way to explain the observed data starting from hypotheses B) or C).
#9657
"This ought to be enough to convince anyone that our intuition cannot be trusted and that heuristics from regular play are irrelevant."
++ Yes, that is right, there exist 5-men, 6-men, 7-men, and 8-men positions won in > 50 moves.
However, there is no proof that any of those positions can be reached from the initial position with best play from both sides.
As always, your thinking is that of a chess player, not a game theorist, and is simply inappropriate. You have this back to front. When you want to PROVE a result (such as solving a game like checkers or chess), you can't rely on the absurd reasoning that if you don't know a possibility is relevant you can ignore it.
There is evidence of the contrary. In ICCF correspondence players may claim a win (even if it exceeds 50 or 75 moves without capture or pawn move) or a draw based on the 7-men endgame table base. Such 7-men endgame table base win claims do not happen, 7-men endgame table base draw claims happen in 10% of drawn games. ICCF games end in draws in 39 moves average, 74% by agreement, 16% by 3-fold repetition, and 10% by 7-men endgame table base draw claim. Over the last 10 years the draw rate has gone up from 63% to 93%.
I thought maybe you were going to eventually get to a point that was relevant, but you didn't. Actually, I am kidding, I never thought you would.
A sample of a few thousand games between imperfect players could never be reliable. It could be a good indication of a statistic influenced by a large fraction of the population. It is obviously worthless as an indication of something that might occur say one in a trillion times. Your claims are even more extreme than that, since you are saying there is no example in what is surely over 10^20 positions relevant to a solution of chess.