What Rating Would Perfect Play Have?

The answer obviously has to be finite, since Stockfish would more than likely score some draws against it, and hence have a non zero score. My guess would be 4000 - 5000. Anyone have any more thoughts on this topic?

Stockfish vs a perfect player in the form of a 32 man EGTB that chooses randomly between moves that maintain a draw (assuming chess is a draw with best play) would probably draw some games.

A perfect player that purposefully chose between the drawing moves to make it harder on stockfish would probably win every game.

For example, even with the limited EGTBs we have, we've already uncovered forced checkmates that are 500+ moves deep. Sure engines are strong, but there are levels of chess that are completely beyond any imaginable future technology.

Still, some top engine programmers have estimated something like 3600... I don't recall exactly, but I remember thinking it was too low.

But, they're actually working in the field so I can't really argue.

Do you have a source for that figure?

Technical stuff is interesting, but isn't really discussed on chess.com forums.

Sorry, I don't quite remember, but my memory is saying from here http://rybkaforum.net/ and also that's where you can get some good feedback on questions like this.

Is it against site guidelines, or do you just mean I won't get much of a response from chess.com users?

Talking about engines isn't against site guidelines, people do it all the time. Chess.com even puts out articles and videos on them and is running that engine competition thing.

This topic was discussed a few years ago at talkchess.com.

But there cannot be a meaningful answer to this question because ratings strongly depend on the size and composition of the player pool.

This is a complicated math question. I believe in pure Elo, that a difference of 800 is a virtual guarantee of a win. In chess I think we typically use 200 as expected and 400 as the extreme likelihood but I'm sure someone else knows the actual numbers. The complication would be a few fold. Does perfect play get a win or a draw? is it only a win with white? The next issue ... the perfect play can only be on relation to the highest non-perfect play... how high was that dudes rating? Perhaps we would assume that perfect would be 200 points higher? Next is this an engine? Would more than one person. have access to perfect? People keep forgetting that Elo is 100% relative. It's not absolute. Your rating is meaningless outside of your player pool.

You talked about another website heresy!!!

For the topic on hand (nothing to do with the message I quoted) I guess that perfect plays rating would depend on the top engine's rating. I think the top engines rating are steadily increasing so I would guess that means that the rating would increase over time. I am presuming that no other engine shows up playing close enough to perfect to get a draw.

This is not a math question but a Philosophical Question: But the answer will actually surprise you. Let us assume the existence of a parallel universe where matter and energy has a different scale of existence. 32 piece End Game Table Base is normal. Now, somehow you have access to this database. How will your elo change?

To answer that, let us understand how Elo rating is calculated in this case.

For Elo greater than 2400, the k-factor will be 10. Similarly, since the "theoretical" rating of this perfect play will be at least 400 elo points greater than a human - a win will increase elo by 0.8 points and a draw will reduce elo by 4.2. Depending on how many chances you get at playing human players, your elo will keep on increasing. In fact, you might even increase it to 6000 or 9000 if you keep beating up human players. 

On the other hand, if you are trapped in an Island with me - who also has access to this tablebase and we both start at 1200 rating, we will continue to be at 1200.

But, should we even be even talking in terms of elo rating for perfect play? Elo ratings compare the relative skill of 2 players. But once you have access to a 32p TB - there is nothing relative about your skill. Assuming that perfect play ends with draw - anyone who plays against you will have only two types of moves ... "!!" (Excellent) or "??" (Blunder). Similarly, any "engine that evaluates with 32p db" will only evaluate in terms of "0.00" or "Mate in x moves". Chess will be reduced to the level of tic-tac-toe, and there will be no question of relative skill. 

Answering your question mathematically, you will lose 4.2 points per draw and gain 0.8 points per win. So until the day an engine is invented that can draw at least 16% of the times against you, your elo rating will slowly creep upto INFINITY. Not 3600, but Infinity. 

But the day an engine is about to draw you at least 16% of the times ... that will be an interesting day.

"What Rating Would You Like to Play?"

I do not know if I understood the question. You're asking, "If someone (or anything) was perfect in chess, what rating would he have?"

I think I did not understand the question, because if I understand it, the answer can be calculated, since at some point it would not matter much if this player won each of the matches he played because he would not win points:

http://www.computerchess.org.uk/ccrl/4040/

In other words, rating is related to field. if Carlsen played only in -2400 tournaments he would see his rating decrease: here too it would be possible to calculate where his rating would stabilize.

If Carlsen were to only play in 1800 tournaments, He will lose 9.2 elo points when he loses, he will lose 4.2 elo points when he draws and he will win 0.8 points if he wins. 

So after 1000 games, if he wins 990 times, draws 9 times and loses 1  time, his rating will have increased by 792-37.8-9.2 i.e. 745 points. 

If he wins 950 times, draws 45 times, and loses 5 times - his rating will have increased by 760 - 189 - 46 = 555 points. 

Perfect!!!

Great post. Not much to add to that. Never seen such a satisfying answer to such a vague question so fast.

Okay then. Let's assume a player pool of several engines that go smoothly from Stockfish 9 to the 32-man tablebase. Does the question become more meaningful then or does it still lead into a brick wall?

That's what I thought: it can be calculated. But against players -2400 (when I posted I thought of putting -2600, but -2400 I found enough) it would have a higher draw rate. Carlsen would be risking something like 4 to 1 in a sense where his actual strength (regardless of rating system) might not be 4 to 1 against a 2400 player. A match between Carlsen and a 2500 player of 100 games Carlsen would have to win 76 games and draw 24 (if I'm calculating right) to increase 0.8.

Thanks for an interesting source.

This discussion is good because it shows that the strength of a player when measured in rating points is relative to the field. I once was third in a tournament where I draw against a first place and won the second place, I felt that it was somehow unfair. I remember that I was not bored because I did not win the tournament, but for not having stayed at least second, lol

Well... do you really understand how the ELO rating is calculated?

You have to take into account several things like player pool(s) and time controls, else any 3 or 4 digit number you have just spelled is just thin air.

Carlsen is currently the World Champion, rated at 2835.

Dronov is the current correspondence WC rated at 2672.

Stockfish is the TCEC champion rated at something like 2460 or so.

The above numbers are just statistics related to ceratin player pools. But they mean nothing more than that: The correspondence WC will win Stockfish very easily under any "daily" time control in a match, let alone the usual 40 days/ 10 moves used in correspondence. Actually I think that Stockfish will fail to win a single game, even in a long match.

Don't ask!

It might spawn another 1000 post topic.

We'll have every genius in chess.com tell us about how their definition of perfect play is correct and everyone else is dumb.