Winnable is not the same as easy
Every game dealt on Fair Solitaire has been won by a solver before it reaches you. That is a promise about the deal, not about how hard the game will be. A winnable game can still beat you, and in some of the games here it often will. This page sets out what the promise covers, what it leaves out, and what the game gives you to close the gap.
What "winnable" means here
A solver is a program that plays a deal by trying moves, backing up when it gets stuck and remembering the positions it has already seen. It keeps going until every card is where it belongs or it runs out of options. When it wins, it writes down every move.
The game then replays that list one move at a time, with the same code that checks your own moves. Only a deal whose replay ends in a win joins the list that New game deals from, and if you type the number of a deal that is not on the list, you get a winnable game instead. FreeCell is the one game we did not check ourselves: other solvers went through the first million Microsoft deals years ago and found eight that cannot be won. Those eight are never dealt here.
So "winnable" means one precise thing: from the first card, at least one order of play reaches the end. It does not say how many such orders there are, how hard the right one is to find, or whether you could find it with what you can see. That last point matters most. For the very first move, which first moves still win counts how many of them keep the way open, game by game.
The solver sees every card. You may not.
In some of the games here, cards are hidden from you. Our solver is allowed to see all of them.
- Klondike deals 21 of its 52 cards face down in the columns. The 24 stock cards are face down too, but you can turn through them as often as you like, so the columns hold the real secrets.
- Spider deals 44 of its 104 cards face down in the columns, and the 50 stock cards arrive in five rows of ten that you see only as they land.
- TriPeaks hides 18 of its 28 peak cards, and its 23 stock cards go through once.
- Golf deals all 35 table cards face up, but its 16 stock cards come one at a time, in order, with no second pass.
- Pyramid shows the whole pyramid, but on the first of its three passes you do not know the order of the 24 stock cards. After that they come round in the same order, and nothing is hidden.
In these games, "winnable" means a way through exists with the cards lying exactly as they lie. The solver can plan around a King hidden deep in the seventh column, because it knows the King is there. You find out when you reach it. A choice that is sensible on everything you can see can still be the wrong one for what you cannot.
FreeCell, Baker's Game, Seahaven Towers and Eight Off are different. Every card is face up from the first move; Seahaven Towers and Eight Off start a few of them in the free cells, face up too. What the solver saw, you see. A lost game in that family is a problem of planning, not of missing information.
How wide the gap can be
The clearest measurement we have comes from Golf. To test the strategy advice on the Golf page, we wrote simple players that cannot see the stock, just as you cannot, and let each one play every winnable deal among the first 20,000. Every deal they played had a way through.
- Playing the first card that fits cleared 43 of them.
- Playing the longest run in sight cleared 150.
- Taking from the tallest columns when two runs were equal raised that to 252.
- Also playing a King or an Ace the moment it fits lifted it to 334, the best of all.
The best of them cleared about one winnable deal in sixteen. These are simple rules, not people, so they say nothing exact about how often a person wins. But none of the deals they lost was impossible. The gap comes from the stock. Golf gives you one pass through 16 cards you have not seen, and even with their order known, the typical winnable deal needs 15 of them; almost one in three needs all 16. The solver knew the order. The players did not.
Klondike shows the same thing in a different way. With 21 cards face down, choosing a column to dig into is often a guess, and as the Klondike page puts it, a wrong guess about which column to dig into can block a game that was winnable from the start. The deal was fair. The information was not equal.
The solver's way is one way, not the best
The solver's games are not a picture of perfect play either. Our solvers stop at the first way through they find, and none of them looks for the shortest. The game pages say so wherever they describe the solver's games.
- In FreeCell, its wins on deals 1 to 3,000 took a median of 59 moves, from 30 to 140, and the page notes that a careful player can do better.
- In Klondike, its solutions to the first thousand winnable deals used between 61 and 103 card moves, 84 on average, plus about 90 turns of the stock.
- In one-suit Spider, the first 500 winnable deals took 108 card moves in the median game, plus the five deals.
- In Pyramid, the median game took 51 moves, from 29 to 69.
- In Golf, a separate search found the fewest stock cards each deal needs. The solver matched it in 299 of 500 games and turned 0.9 more on average.
TriPeaks shows how narrow the solver's own way through can be. Searching each of the first 500 winnable deals to every position shows what a perfect player, one who knew every card, could do: keep a median of 9 stock cards unturned, where the solver kept 2. The solver needed the last stock card in 105 of those 500 games; a perfect player would have needed it in 10. A win that comes down to the last stock card is in good company.
What the game gives you
Three tools help close the gap between a deal that can be won and a game that you win.
Undo, as a scout
Undo has no limit, and it takes back anything: a card moved, a stock card turned in Golf, TriPeaks or Pyramid, a row dealt in Spider. That makes it a way to look. In Klondike you can make a move, see what the face-down card is, and undo it if it was not what you needed. In Pyramid you can draw through the stock, note the order and undo, and you know what the solver knew from the start. The Golf and TriPeaks pages suggest the same, and add that whether it is fair is up to you. In FreeCell and its relatives nothing is hidden, so trying a line and taking it back is simply how the game is played.
Hint, which says when the game is lost
Hint asks the solver for a next move on a way to win. Like the solver, it knows the hidden cards and the order of the stock, so its move can depend on cards you have not seen. When it finds no way to win from where you are, a box opens. If the search proved it, the box is headed "No way to win from here" and says "This position cannot be won any more." If the search gave up before settling it, it is headed "No way found from here" and says "We could not find a way to win from this position."
Either way, it then says how many moves you would have to take back to reach the newest position that can still be won, with a button to go back that far, one to restart and one to keep playing. If the only such position is the start, it suggests a restart instead, because the game can be won from the start. The second message is there because a solver has limits too: in FreeCell, Hint gives up after 200,000 positions, and on about one deal in 170 it finds no way even from the start, though the deal can still be won.
If you run out of moves altogether, the game says so, adds that the game can still be won from the start, and offers to find where it went wrong.
Restart
Restart deals the same game again. Whatever you learned the first time stays with you: where the Aces were, which column hid the King, what came out of the stock. In a game with hidden cards, a second attempt is a different game for you, even though the cards are the same. Every game also has a number, so you can come back to a deal later or share it.
Why the promise still matters
If winnable does not mean easy, why make the promise at all? Because of what a loss means without it. Under the rules here, three Golf deals in four cannot be won however well you play. About a third of Pyramid deals cannot be cleared. Roughly one Klondike deal in ten is impossible turning one card, and about one in five turning three. In Baker's Game it is about one in four. A site that deals every shuffle deals those games too, and when you lose one, nothing tells you whether the deal could have been won.
Here that question always has the same answer. When you lose, the way through was there. Hint can tell you how far back you left it, Undo can take you there, and Restart lets you play the whole game again with what you now know. A site that deals impossible games cannot offer that, because for those games there is nothing to go back to.
What the promise does not do is play the game for you. In FreeCell and its relatives you work with exactly what the solver had. In Klondike, Spider, TriPeaks, Golf and Pyramid you work with less, and some of your losses will come from that. They are still losses with a way through, and that is the part we can check before a game reaches you.
Read more: how many games can be won, game by game, and why we made Fair Solitaire.