How the hell is this endgame WINNING for White????

the endgame table base shows this endgame to be winning?? I am highly skeptical that this is a bug, rather this is an actually winning endgame for white, but how???? white only has 2 knights remaining, which is well known that they cannot force mate, but give black 2 pawns and white is suddenly winning?? I assume it has something to do with stalemate, in this case black will have pawns to move, whereas 2 knights vs a lone king would end in a stalemate, but how exactly is white supposed to stalemate blacks king?? is there some known technique for 2N vs P/2P??? or is it only possible in very rare situations??

2 Horseys vs 1 pawn= winning! Here is the secret technique!
lichess.org/forum/lichess-feedback/two-knights-mate-lesson#1

Two knights alone cannot force a checkmate but two knights against a pawn which is not advanced enough (how much depends on the file) can force a checkmate. But it's extremely difficult, even harder than a checkmate with bishop and knight. All I know is that the trick is that you could actually force the defending player into a checkmate position but you would have to go through a position which is a stalemate - and that's where the opponent's pawn comes handy.

IIRC there was even a period of time when there was an exception that the 50 moves limit was extended to 100 moves in these positions. (Not the case any more.)

endgame table bases do not have bugs due to method they are compiled. Or they would have a lots of bugs. Building table is pretty simple backtracking algorithm. For sheer fun of it i clicked trough the table 1st move and mate happens at move 116.

For a win for the knights (ignoring the 50 move rule) the pawn needs to be behind the "Troitsky Line" and securely blockaded by a knight. For an e-pawn as here, the pawn needs to be no further advanced than e4. This did not actually occur in your posted game, because with the entry into the K+N+N vs K+P endgame on move 55 the pawn immediately advanced to e4 and was not blockaded there.

So that particular endgame was in fact drawn.

no it is not a draw. higly academic issued but 7 piece table base says it is winnable it really is. In practice far and few between can really win this

@petri999 said in #4:
> endgame table bases do not have bugs due to method they are compiled. Or they would have a lots of bugs. Building table is pretty simple backtracking algorithm. For sheer fun of it i clicked trough the table 1st move and mate happens at move 116.

I assume you must have chosen a slightly different starting position than the positions which occurred in the given game, maybe with the pawn further back? The position at move 55 on entry into the K+N+N vs K endgame appears to me to be drawn.

@petri999 said in #6:
> no it is not a draw. higly academic issued but 7 piece table base says it is winnable it really is. In practice far and few between can really win this

That's very interesting, because the pawn is too far advanced for conventional theory to indicate a win. On e3 it's in front of the Troitsky line. Perhaps the conventional theory is wrong?

8/2Nr4/4Kp2/4pN2/4Pk2/8/8/8 w - - 0 53
the position I saw on boad and table base gives Kxd7 as winning move. in actual game hitting win Kn is drawn.

Conventional theory has shortcoming that it compiled by human analysis which by necessity means it is limited. It must assume some fixed procedure to be analyzable. While backtracking from end position takes less brain but is guaranteed to come to solution given enough of storage space and cpu.