How close is Stockfish 15 to perfect play?

ELO rating does not ensure perfect chess. it means top of a pool. the whole pool could be stuck in some corner of chess space and still compete there.
when asking about perfect chess, one has to consider the whole of chess space.
need something else than ELO, or have some mechanism to ensure coverage of possibililities.

in all of this is the belief in one best decision for all the positions as part of perfect chess, because engines have been built to always chose a best move, even when in front of flat profiles. look at the TBs. or many cases where there are literal ex-aequo solutions that are perfect within human calculation reach. it is not likely that this kind of non-unique perfect solution be restricted to short sequences. We should not use engines as absolute measures, they are rated with same basic flaws as we are.

it looks actually very probable to me. mate or draw as best outcome is already possible with many equal depth paths for many positions. In chess there are three outcomes, not three final positions. this is not an annihilation game. or other reduced terminal set of configuration type of game.

given the wikipedia definintioins. TB is strong solve (for its initial positionS).
and the one solution for one color is (forgot) weaker, but not weakest, which is the non-constructive proof of existence of solution. (yet my preferred way to think, i am a weakling, I love to imagine stuff i can't construct).

If perfect play is the ability to have the optimal result from any position, there is still a way to go. But if it's just never losing from the start position no matter what the opponent does, strong correspondence players playing conservatively using their opening book, Stockfish and tablebases are probably already there.

Historically, weaker engines could still exploit holes to sometimes beat an engine 50-100 elo stronger in a position, but this isn't the case anymore. Stockfish is insanely strong.

I just meant ELO as a proxy for how chess engines are improving (and if they are still improving). If they stop improving their ELO I'd guess that means we aren't progressing towards the eventual goal of 'perfect play' UNLESS all games have a predictable outcome at that point.

Anyway, i'm curious whether such a plot exists vs time.

Stockfish hasn't stopped gaining elo yet:

user-images.githubusercontent.com/64992190/174315817-6c526ca4-6d26-4322-8775-561e10feab4d.png

But that's at bullet TC on a single slow CPU core. On longer time controls, or with more threads, the gains are compressed.