How close is Stockfish 15 to perfect play?

@Akbar2thegreat Clearly you know what you are talking about, so I'm not going to argue.
I believe scientifically it is not impossible, but I doubt anyone would try to do it - it ruins the whole human element of chess and why we enjoy it.

@continuum12
I was telling my view on perfect play and solving chess and also answered your question at the end.
Currently we have 7 piece tablebase and now 8 piece endings are being tried out for solving. Since number of chess positions (10^120) are way more than number of atoms in observable universe (10^120), solving chess would require solving all 32 piece positions, that time would be very large.
Well, these discussions don't yield anything but in all, it is very unlikely that chess will ever be solved despite any existence of perfect play.

@Akbar2thegreat I get what you're saying - "solving chess" would take a 32-piece (at this point not a) endgame tablebase, which would either be impossible with today's computers or take a very long time.

I'm thinking now, what would 32-piece tablebase look like? Forced win with White? Forced draw? Proving some openings are better than others?

@continuum12
Well, result of perfect chess game will depend on first move.
And first move advantage is not enough as many experts believe.
Majority of chess theorists believe that chess is draw with perfect play.
Even I believe so.
It's a long debatable question in history of chess but for answer it needs to be solved which it wouldn't be solved anywhere in near future.
Also, read:
en.m.wikipedia.org/wiki/First-move_advantage_in_chess

@Dallasisanortard2 @Akbar2thegreat
So what would happen if Stockfish 15 played itself? Would it result in a draw every game? Or would white occasionally win?

@Epop44
Among equal strength chess engines of same version and CPU, the game ends in a draw.
Also, better ask Stockfish itself!

did not read the posts before. but my answer is we will never know. if all measures used to date stay the same.

@Akbar2thegreat @Dallasisanortard2 It's reassuring to consider, that relatively small- modest* changes to Chess play would result in complete overhauls of engine tactics. Variations like:
(1)If a player elects to castle, their first pawn promotion can only be a to a knight or bishop.
(2) When Promoting, the new piece (Bishop, Knight, Queen Rook) is selected by the opponent)
(3)If a player wishes to move their black squared bishop onto a white square (or vice versa), player, can use a turn and move it on the square directly to the right or left, (so bishops would no longer be confined to a single color)

How would the engines cope?

*Compared to variations like Chess960

@Epop44
Engines aren't made for variations.
They were made to help in analysing highly complex standard chess.
And because of it's complexity, variants were made by players like Alekhine, Capablanca, Fischer, etc.

I would think that engine based on variation exploration in legal space would be made for variants that keep the core mobility rules.

so I would expect a pure legal search min max (unbiased and complete, together with an inbiased and complete static evaluation for unbiased and complete subset of explored nodes to be evaluated) to be able to withstand the initial condition change only difference of the 960 variant (or more precisely, extension) of stantdard chess.

be careful to use variation for legal space exploration, and variant for rule set exploration. it helps with faster reading. some might also get stuck on that.

edit: possibly replace "complete" with "covering" (depends on smoothness of embedding space from chess encoding).