Case Study: 7 Knights vs 3 Queens is a draw!

@MARattigan

So, you agree you are mad! (jk)

Well, I prefer mad to odious.

In short: I was justifying the leveling effect with an example. And since everything depends on position, not every position shows leveling effect to that result producing extent.

You were justifying nothing. Your method is flawed.

Your definition of "leveling effect" doesn't correspond with Betza's in any case.  

And the position I initially posted goes well beyond Stockfish.

As I already pointed out here. So why are you trying to use Stockfish to prove your position is a draw?

Have you ever seen when does Stockfish suck in puzzles? It fails to understand obvious human moves. Hence it fails in analysing fortress and closed positions. Exactly the reason why Stockfish initially shows the position I posted as winning for Queen side. But on doing deep analysis upto high depth and following top moves, the computer analysis shows that without any bad move from Queen side the game ends in draw.

No it doesn't show that.

Several reasons.

You say "With computer analysis and also from analysis of Stockfish 13 NNUE, I found that without 50 move rule, the knight side can still force a draw by forming some particular positions on board where Queens can't do anything".

The fact is that Stockfish has no options to play without the 50 move rule. It evaluates positions according to the 50 move rule ply count. If your position is evaluated with different ply counts you get different results. For SF14+NNUE at depth 20 you get:

Ply count 0:  -5.50

Ply count 50: -1.93 (half of 50 move rule)

Ply count 75: 0.00 (three quarters of 50 move rule)

This results in play specifically tailored to the 50 move rule. It's likely that your position is a draw with the 50 move rule in effect (only likely not proven) but nothing can be said about the result  when the rule is not in effect.

You also don't have the computing power to do "deep analysis upto high depth". The depth of search is a logarithmic function of time or computer speed. 

The 1 queen against 2 knights endgame may need 72 moves to mate. How many moves could 3 queens against 7 knights need?

If Haworth's law holds good up to 12 men it would suggest a longest forced mate with twelve men on the board around 25,000 moves. Your position could well be a mate in more than 1000 moves - we'll never know.

You also fall into the trap of assuming that a deeper search will produce more accurate evaluations. You should read some papers on minimax pathology. Here is D.S.Nau's initial paper on the subject.

In this post on a different thread I gave a series of games where SF14 was similarly out of its depth, but in the two knights vs pawn endgame, with fixed times per move ranging from 1 sec to 37 min. I then went through the games using the Syzygy site to identify the blunders (moves where the SF evaluation was so incorrect as to alter it's theoretical result for the worse).

This is a Wolfram linear regression graph of the results when the blunders are identified in the FIDE basic rules game (no 50 move rule).

You will note that the linear regression line shows a general increase in the blunder rate as the think time allocated is increased.

Whether or not the same would be true in the 3 queens v 7 knights endgame I don't know.

Neither do you. 

You should also see 'Difficult for Humans, Impossible for Computers' playlist of Chess with Suren.

@darkunorthodox88

Exactly I also read that somewhere but unable to locate the source right now. Also better tell this to @)MARattigan who doesn't believe in such things and doesn't use simple logic.

You're very good at deciding what I think and know on the basis of nothing.

I have in fact read Betza's online postings on the subject. (You've apparently based your information only on a one line misinterpretation of the same appearing in Wikipaedia.) 

He does indeed discuss the fact that the tendency of multiple pieces to block the paths of other pieces should reduce the valuations, based purely on mobility, of the higher valued pieces. It makes a lot of sense. He also talks about redundancy as does @darkunorthodox88. That this should have a similar effect also makes a lot of sense.

He doesn't call either 'levelling' (or 'leveling') and he doesn't say that it results in 3 queens either losing (Wikipaedia) or drawing (your assertion) against 7 knights.

I mostly use simple common sense.

I posted here five randomly generated 3 queens v 7 knights positions.

Here are another five.

That is ten altogether. Nine are wins. The tenth is beyond my ken and that of SF14 (and, I would suggest, your own). It could be a win, as it is when SF14 plays it against itself, or it could be a draw.

That should be enough to persuade the averagely intelligent beetle to dismiss your assertion "7 Knights vs 3 Queens is a draw!" as nonsense.
 

@MARattigan

You aren't still getting it. As don't try to play with words.

I was only explaining why 3 Queens 'may' badly lose due to leveling effect.

Except you say your position is a draw, not a loss, and you don't show any connection between the Wikipaedia definition of "levelling" that you use and why it should be a draw.

You don't in any case show it is a draw.

Ask any chess theorist who is also a logician, and they will tell you that 3 Queens vs 7 Knights a draw with perfect play though not proven.

I don't ask any chess theorist; I work it out for myself. It's generally a win for the queens.

In fact, generally a quick win for the queens.