Is the "Wolf and Sheep" chess variant a forced win for white?
14
Ṁ5496
2040
98.1%
chance

"Wolf and Sheep" is a simple variant of chess that's useful for teaching basic pawn strategies and how to plan several moves ahead. (Well, there are a few different variants with that name, but I'm referring to a specific one.)

White begins with only its 8 pawns. Black begins with only its queen. All pieces move normally. White wins if it gets a pawn to the 8th rank or captures black's queen. Black wins if it captures all the pawns.

Resolves to YES if it's a forced win for white, NO if it's a forced win for black, 50% if it's a draw.

Get Ṁ1,000 play money
Sort by:

Is it a draw and resolves 50% if the winner could be either and depends on which side uses better strategies like in regular chess?

edit: shoot i was running with the wrong side as black. ignore me lol

Huh. I analyzed with stockfish 16, and it claimed that the position was a win for black...

@Qoiuoiuoiu Note that in the wolf and sheep variant described in this market, there are no kings on the board. It's just whether the queen can capture all the pawns before one promotes.

predicts NO

@BoltonBailey Ah thanks, I didn't see that.

predicts NO

@BoltonBailey I analyzed on https://fairyground.vercel.app/advanced.html, but I easily beat the computer opponent, even when it had a ton of time for analysis...

predicts YES

What proof is needed to resolve this? I ran Fairy-Stockfish on my laptop and found a forced win. You can even run it in the browser.

Since white starts, the move d3 can be input as a staring point for a search for forced-win. From this, Fairy-Stockfish eventually finds the evaluation to be mate-in-16. Which means it's found a forced win in at least 16 moves, it continues searching the game tree for faster forced sequences, but that should not be relevant to resolve this market.

Example sequences that engine outputs (given the move d2d3 as the first move by white):

d8d4 a2a4 d4a4 c2c4 a4b4 f2f4 b4f8 b2b4 f8f4 b4b5 f4f2 h2h4 f2e3 g2g4 e3a7 g4g5 a7g1 h4h5 g1g5 b5b6 g5g8 h5h6 g8h7 c4c5 h7f7 c5c6 f7g8 h6h7 g8h7 c6c7 h7h1 c7c8q
d8d4 a2a4 d4a4 c2c4 a4b4 f2f4 b4f8 b2b4 f8f4 b4b5 f4f2 h2h4 f2e3 g2g4 e3a7 g4g5 a7g1 h4h5 g1g5 b5b6 g5g8 h5h6 g8h7 c4c5 h7f7 c5c6 f7g8 h6h7 g8h7 b6b7 h7h1 b7b8q
d8d4 a2a4 d4b2 e2e4 b2c2 f2f4 c2d3 e4e5 d3f5 a4a5 f5f4 e5e6 f4e3 a5a6 e3a3 h2h4 a3e7 h4h5 e7c7 g2g4 c7g7 h5h6 g7e7 g4g5 e7c7 g5g6 c7b8 e6e7 b8e8 a6a7 e8e7 a7a8q

It's easy to see why it's a forced win. White can quickly get 2 pawns up to rank 6, which means black queen can't capture these pawns without seeing the other pawn advance to rank 7 without capture next turn - which is a loss. White can then use this to get even more pawns to rank 6, until it has enough to overwhelm the black queen, resulting in a win for white.

Here is the variants.ini file:

[wolfandsheep]

pawn = p

queen = q

startFen = 3q4/8/8/8/8/8/PPPPPPPP/8 w - - 0 1

promotionPieceTypes = q

castling = false

extinctionValue = loss

extinctionPieceTypes = *

flagPiece = q

whiteFlag = *8

@daniel This looks promising! I tried playing around with it myself here, but couldn't quite get it to work. (I'm not familiar with this program, so I'm unsure how to interpret what it tells me.)

How long did it take to find the forced win when you ran it?

predicts YES

@IsaacKing It didn't take more than maybe 10 minutes or so on my machine. But after inputting the variants.ini, and selecting wolfandsheep in the dropdown, put d3 on the board, then press "Go". The engine runs to the left, and outputs commands in UCI format (https://en.wikipedia.org/wiki/Universal_Chess_Interface).

The engine outputs "info" commands, and the "seldepth" is the deepest search for the best move. There is a "score" field alongside that either shows a centipawn ("cp") score that is the engine's evaluation of the position, or it shows how many moves for a forced mate ("mate"). Since we're evaluating black's position, negative values mean white is winning.

At depth 37 or so if you run the same version as me, you should see it has searched the game tree and found the position to be a forced win for white with best-play (mate in 16). The game tree is of course enormous, but if we trust Fairy-Stockfish to be programmed correctly, the game you described is a forced win for white.

predicts YES

@IsaacKing Actually can run the engine on starting position as well, but it's faster after d3 of course.

@daniel Oh I figured out the problem. Simply importing the .ini file is not sufficient; I then have to select the variant from the dropdown afterwards.

Well the primarily thing I've learned is that I'm terrible at Wolf and Sheep. It consistently beats me no matter which color I play.

I don't think I want to resolve this market quite yet, since there's still a reasonable chance that you made a mistake or Fairy Stockfish contains a bug. If a second person uses a separate method to analyze the game and gets the same result, I'll consider that sufficient. I'll even allow it to be a non-proof that's highly suggestive, such as some grandmasters playing against each other and switching off sides, and finding that white always wins.

predicts YES

@IsaacKing You can verify that the rule-set corresponds to the rules you stated, both on the website and in the .ini file:
- Two pieces from chess, pawns and queens ("pawn" & "queen")
- They start in the starting position of normal chess, with white to move ("startFen")
- If black or white loses all their pieces, it's a loss ("extinctionValue" & "extinctionPieceTypes")
- Pawns always promotes to queens ("promotionPieceTypes"), and if white has a queen ("flagPiece") on any flag squares ("whiteFlag"), then white wins. The flag squares are all files of the 8th rank (="*8" which means "a8" or "b8", ... or "h8"). All in all, if white ever gets a pawn to 8th rank, white wins.

Other than this, it's chess rules, so stalemate will still result in a draw, for example.

The probability Fairy-Stockfish has a bug I'd consider very small, it'd have to be in some core component in that case.

I think your resolution criteria are overkill, clearly it's a forced win for white. The likelihood someone else will study this is small, since it's a unknown variant that can't be found online and is not documented anyhwere.

predicts NO

@daniel You're sure that it's actually doing an exhaustive search and there's no heuristic tree pruning?

predicts YES

@IsaacKing It uses alpha-beta pruning. If the evaluation comes out to mate-in-X it's proven to be forced.

Fairy-Stockfish finds win-in-17 for white with these rules.

Made a bet on the basis of this analysis, but upon reflection the fact that white has no king to check and wins if it promotes (even if the promoting square is controlled by black) are major factors in white's favor.

@BoltonBailey Yeah, that's an interesting analysis, but I don't think it says much about this game. Wolf and Sheep tends to be much more of a race, since there isn't a king to defend the pawns from the queen.