Which engine is best in winning drawn tablebases endgames?

[Uri Blass]

it is possible to win drawn tablebases endgames against weak engines that do not use tablebases but I suspect stockfish is not the best in doing it because it is too smart and does not know to go for drawn positions when the opponent may go wrong because it evaluate them often as 0.00 even without tablebases.

[Joerg Oster]

it is possible to win drawn tablebases endgames against weak engines that do not use tablebases but I suspect stockfish is not the best in doing it because it is too smart and does not know to go for drawn positions when the opponent may go wrong because it evaluate them often as 0.00 even without tablebases.

Wrong question, imho. You can't win a drawn endgame.
A weaker engine may eventually blunder away a drawn endgame, but that's not quite the same.

[Uri Blass]

it is possible to win drawn tablebases endgames against weak engines that do not use tablebases but I suspect stockfish is not the best in doing it because it is too smart and does not know to go for drawn positions when the opponent may go wrong because it evaluate them often as 0.00 even without tablebases.

Wrong question, imho. You can't win a drawn endgame.
A weaker engine may eventually blunder away a drawn endgame, but that's not quite the same.

You can help the weaker engine to blunder and from my point of view making moves that practically give better chances to win against weak engines (and probably also against humans) is better.

When I analyze my games with chess engines I want to know where I could make moves that give me better practical chances(engines that give 0.00 for different drawn positions even if one of them has better practical chances to win do not help and I want to know which engine to use to analyze my games when I am not sure if stockfish is the best)

[syzygy]

it is possible to win drawn tablebases endgames against weak engines that do not use tablebases but I suspect stockfish is not the best in doing it because it is too smart and does not know to go for drawn positions when the opponent may go wrong because it evaluate them often as 0.00 even without tablebases.

Wrong question, imho. You can't win a drawn endgame.
A weaker engine may eventually blunder away a drawn endgame, but that's not quite the same.

Same thing for chess from the starting position. Still we call it winning.

The goal of chess is to lure your opponent into making a blunder.

[brianr]

it is possible to win drawn tablebases endgames against weak engines that do not use tablebases but I suspect stockfish is not the best in doing it because it is too smart and does not know to go for drawn positions when the opponent may go wrong because it evaluate them often as 0.00 even without tablebases.

This would be difficult to do with engine v engine matches.
However, for engine v human, Crafty had "swindle mode" to do this, IIRC

[syzygy]

it is possible to win drawn tablebases endgames against weak engines that do not use tablebases but I suspect stockfish is not the best in doing it because it is too smart and does not know to go for drawn positions when the opponent may go wrong because it evaluate them often as 0.00 even without tablebases.

This would be difficult to do with engine v engine matches.
However, for engine v human, Crafty had "swindle mode" to do this, IIRC

Stockfish does something very similar (last time I looked). If the position on the board is a tablebase draw or loss, it will switch to searching without probing the tablebases. (In case a of a draw, it will not search any losing moves in the root position.)

Uri's point is that, even in swindle mode, i.e. even if Stockfish does not probe the tablebases, it will usually know that the position is drawn and therefore will not be able to distinguish between drawing moves that make it difficult for a fallible opponent to hold the draw and drawing moves that make it easy for a fallible opponent to hold the draw.

In a way Uri's question has nothing to do with tablebases. If an engine has "perfect" search and evaluation, that can hurt the engine when playing against fallible opponents. Suppose a future engine is able to see that 1.a3 and 1.e4 are both a draw, then why would it pick 1.e4 over 1.a3? But of course 1.e4 will (very likely) give the engine better winning chances against a fallible opponent than 1.a3.

In practice the iterative search of Stockfish will probably tend to prefer "subjectively" better moves over "subjectively" worse moves, even if the moves are all draws and therefore "objectively" equally good.