Pawn promotion to Bishop (NP vs N) ( my puzzle)

Did you liked? or it's easy puzzle?

Syzygy can hold position but engines not

Notice I won aganist Stockfish 11

with black pieces ( get winning position but game is ended draw with white pieces)

with Traxlercounterattack ( a lot of engines play Nxf7??  (without book) )

but this position is not starting position 

You have made a mistaken analysis I think? Stockfish says this is a draw.

Even after you get the bishop

K+N+B vs K+N is a draw

@antisunechess

K+N+N fully draw

K+N+Bvs  K+N

possible trap knight

you can use 

Syzygy or find some  basic puzzles ( K+N+B vs +K+N)

https://www.shredderchess.com/online-chess/online-databases/endgame-database.html

use this website

Then I guess stockfish is not so good after all. Maybe Stockfish 12 will fix this?

@antisunechess if you use

Syzygy  Engine says draw because 50 move rule

I think engine can't defend this position 

chess engines have depth limit 

I think no

K+N+N vs K+N 

rarely white wins game ( ıf black sacrifice knight and don't made blunder ( mate in 1)

it's draw)

Syzygy says it is a cursed win which means that it draws if the 50M rule applies. However, it does not if you present this diagram as a regular endgame study. 50M does not apply to checkmate problems and endgame studies, only to chess games and retrograde problems.

In other words, it is a correct endgame study with a win for white!

A human is unlikely to win this endgame against Stockfish since we are much worse at calculating than this engine. We won't beat Syzygy either if only it is set to delay defeat as long as possible. It is simply too difficult for us. Note that Stockfish commonly uses game rules including 50M. There is no point in playing an endgame study against an engine with different rules.

@Arisktotle true

Black Nd7 forced black knight would be trapped ( after 50-60 mvoes)

I think we can win again stockfish 11 ıf we play perfectly because Stockfish 11 can't find best defense

moves allways

look like Black knight can be trapped very easy ( because black king is so far )

but it's not easy at  N+B 

need technique to win

it's very rare position

Another good example of promotion to bishop ( same idea here) still cursed win here

But this position is not

not finished

Precisely! But we humans can't play perfectly, not even close. What you are really saying is that Syzygy can beat Stockfish 11 because it is perfect and Stockfish is not. I have no argument with that! And keep in mind that Stockfish plays with 50M. I don't know how that affects the Stockfish evaluation scores and its move selection.

I already found your diagram #10 by playing around with Syzygy. It is the worst position for the white knight if I recall well. But your original composition has the best starting sequence because the move 2. Nf7! is very powerful!

Note; it is not a disadvantage to have a cursed win but an advantage. It is always good when you can demonstrate something in a composition outside the scope of normal game play! The 50M rule is not native to chess but is artificially added to control game duration. Why would anybody want it in an endgame composition?

But it isn't. It's set neither to win as quickly as possible nor to lose as slowly as possible. In fact if it's in a losing position it can give the impression that it's deliberately throwing the game (and in a winning position it will sometimes make apparently ludicrous moves).

What you can say is that with the 50 move rule in effect an engine using Syzygy will successfully win a winning position, successfully at least draw a drawn position and successfully at least lose a losing position. It takes not too much notice of the number of moves.

E.g. downloaded from the Syzygy site this White to win in 18 (my comments in brackets).

Note that Sysygy with the white hat successfully won and Syzygy with the black hat successfully lost, but if Syzygy is in a losing position it won't necessarily give you a hard time. Note also that the 50 move rule is not relevant in this example.

If Syzygy is in a winning position then the story is different. No matter whether the 50 move rule is or is not in force and no matter how accurately you play, you will lose. Here it is crushing Nalimov from the same mate in 18 position:

Of course, when I say you will lose, it could be on time.

I know! It isn't because it isn't set up to behave as a decent opponent for human problem solvers. In the context of the discussion I only tried to convey that man controlled Syzygy is superior to Stockfish and human brains in 50+ endgames by just manual selections of maximum delays for the defense. Any engine permitted to incorporate the Syzygy features can easily do the same without human assistance. I didn't intend to describe the characteristics of the Syzygy interface itself, only the power and scope of the information it yields for use in endgames. I would have chosen a different formulation had I written the reply for you! 

I think the endgame is an unfortunate context in which to extol the virtues of Stockfish.

A human player who has studied a particular endgame with a small number of pieces will, in fact, generally out-perform Stockfish if it has no EGTB to look up the answers. E.g. here is one I prepared earlier where I check out SF8 on a random mate in 50 KNNKP position (at G120+10)

I stopped there because my king was in the drawing zone and SF8 had already been drawn for three moves anyway. 

Of course SF appears to perform infallibly in endgames that one hasn't studied whether these are just a few pieces or the full 32 and a human can't generally get a complete grasp of endgames with more than a few pieces, so SF appears better in the opening and middlegame (though in reality one would expect it is much worse).

Many of the shorter endgames have fixed patterns and strategies which can be picked up by humans but not by SF. 2N vs P is one of the longer ones which is "learnable" though it requires a lot of study. But how do you "learn" endgames of the type posted by the OP? Of course, you must cut off the lonely knight but such is not of great help in itself. As soon as endgames go beyond the point of being "learnable" (by a human, not a neural computer network) then the precise calculation skills of SF are of more use than a few strategic generalities.

Of course, there is a complete new category of issues connected to having "infinite time and opportunity" without 50M or repetition rules. Instead of learning "how to win" you can concentrate on learning "how not to throw away a win". If you can do the latter, you will ultimately win by making just random attempts within the boundaries of "not throwing away the win". Still, I doubt you can define rules for "not throwing away the win" in the OP's endgame.

These things make it hard to compare the efforts of SF and humans. I suggest that your notion of "accuracy" is a fair measuring rod to replace infinite attempts, repetitions and 50M-like limitations. Just look at how many moves the defense holds out in a lost position, how long the winner takes to achieve his objective, and of course, who achieves the perfect (syzygy-wise) score at all. Do not forget to cut off at the "billion move draw line" as explained in my "billions" post a few months ago . You always need a termination point!

In a sense this whole discussion is aging quickly into oblivion. Probably every chess engine will have either an online or an onboard tablebase available in a decade or so. It is just a bit silly to approach "small" endgames by calculation or strategies while tablebases are clearly the better way! For computers that is.

Humans already do of course at http://tb7.chessok.com/probe or https://syzygy-tables.info/. On the other hand computers are probably temperamentally better suited to withstanding the boredom of looking all the moves up in a table when practising these endgames than are humans.

I did already have a program that could play endgames such as the one shown without any EGTB a decade or two ago. It was a version of Rybka with an "e" for "endgame" on the end of its version number - unfortunately now trapped in a non functional disc drive. 

I like to occasioally practise against something that will throw in the odd mistake. Mistakes seem to beget mistakes, so I think it's useful.

A human consulting a tablebase during a game is "cheating". An engine consulting a tablebase during a game is "doing its job" 

What do you mean with "without any EGTB"? Did you use a private EGTB or none at all? If not, did you get perfect results? Quickly?

I mean I didn't give Rybka an EGTB. I don't use one, but I do practise against Nalimov. Playing against Rybka I usually played the same position from both ends (I got it before I heard about EGTBs). 

I had to give Rybka at least 20 minutes to play a 30 - 50 move KNNKP position if it was to behave itself, but increasing that didn't give any improvement that I could detect. I generally picked up about five moves on a two way play from a given 30-50 move position. About one in twelve times it made major blunders on those and it always started falling apart beyond 55 moves. (Having since got the regular version, which can't even play KBNK, I strongly suspect that the problem there was not the event horizon, but the 50 move rule.)

As for perfect results I only clock those playing against Nalimov. I used to outplay Rybka in general (specific endgames only of course). And as for speed I certainly played a lot quicker than Rybka, but the games were a lot slower than those against Nalimov and I would be making coffee and watching television  etc. while it thought, so I never got any figures. 

That's on KNNKP positions. I used to practice KBNK against it too (positions more than 18 moves to mate) but I'd only pick up about 2 moves per three games on those. SF8 is about the same strength with the same time control on that ending but SF8 can't play KNNKP endings over about 20 moves at all on any time control I'd like to play it on - or maybe even would live long enough to play it on. (I suspect SF10 might, having played a time constrained version on lichess.)