How winnable is solitaire? The odds of winning 9 games, from 63,000 deals
Some solitaire deals can never be won, however well you play. We ran our own solver on 63,000 numbered deals of 9 solitaire games, the same games you can play on this site, to see how many can be won. FreeCell is almost always winnable, fewer than half of Golf deals are, and at least half of Scorpion deals are impossible.
The short answer
- FreeCell is almost always winnable. Of the 32,000 classic numbered deals, our solver won 31,999 and proved the last one, deal #11982, impossible: it checked every reachable position (61,643 of them) and none leads to a win. That matches what FreeCell players have said for 30 years, now confirmed again by an independent program.
- Golf: fewer than half of deals can be won. We settled all 5,000 Golf deals we tested: 2,254 can be won and 2,746 cannot, so 45.08% are winnable, matching a 10-million-deal university study (45.109%).
- Scorpion: at least half of deals are impossible. We proved 51.37% of 3,000 Scorpion deals impossible (1,541 deals), so at most 48.63% can be won, and our solver won only 12.47%. A longer search on 150 of the undecided deals proved 94 more impossible and won just 1.
- Most deals of Yukon, Pyramid and Canfield can be won, if you could see every card: our solver won at least 68.1% of Yukon deals, 62.83% of Pyramid deals (three passes) and 58.56% of Canfield deals, and proved 3.1%, 10.27% and 8.98% impossible.
- Klondike, the Windows classic, is harder than our solver. It won 68.87% of turn-one deals and 64.27% of turn-three deals. University researchers with a much stronger solver put the true figures at about 90% and 82% (Blake and Gent, University of St Andrews). Our numbers are a floor; theirs are the answer. The same goes for Spider and Forty Thieves.
“Winnable” here always means: can be won with perfect play by someone who can see every card, including the face-down ones. Real players can’t, so they win less often (one skilled player won 189 of 442 Klondike games, 43%, in a test reported by Jupiter Scientific). Researchers use the same convention, so our numbers can be compared with theirs.
Advertisement
The results
| Game (how we play it) | Deals tested | Won by our solver | Proven impossible | Still undecided |
|---|---|---|---|---|
| FreeCell (deals 1–32,000) | 32,000 | 31,999 (99.997%) | 1 (0.003%) | 0 (0%) |
| Golf | 5,000 | 2,254 (45.08%) | 2,746 (54.92%) | 0 (0%) |
| Scorpion | 3,000 | 374 (12.47%) | 1,541 (51.37%) | 1,085 (36.17%) |
| Yukon | 3,000 | 2,043 (68.1%) | 93 (3.1%) | 864 (28.8%) |
| Pyramid (3 passes) | 3,000 | 1,885 (62.83%) | 308 (10.27%) | 807 (26.9%) |
| Canfield (turn 3) | 5,000 | 2,928 (58.56%) | 449 (8.98%) | 1,623 (32.46%) |
| Klondike, turn 1 | 3,000 | 2,066 (68.87%), at least | can’t tell | 934 (31.13%) |
| Klondike, turn 3 | 3,000 | 1,928 (64.27%), at least | can’t tell | 1,072 (35.73%) |
| Spider, 1 suit | 3,000 | 2,501 (83.37%), at least | can’t tell | 499 (16.63%) |
| Spider, 2 suits | 2,000 | 1,509 (75.45%), at least | can’t tell | 491 (24.55%) |
| Spider, 4 suits | 400 | 115 (28.75%), at least | can’t tell | 285 (71.25%) |
| Forty Thieves | 600 | 84 (14%), at least | can’t tell | 516 (86%) |
“Proven impossible” only appears for games where our solver’s search is complete, so that running out of positions is a proof. For Klondike, Spider and Forty Thieves the search takes shortcuts, so it can win deals but never prove one lost: their “won” column is a minimum, and the true share is higher. Klondike’s turn-one floor rises to 75.63% if you count the 255 layouts our solver won only at turn three, since a turn-one player can copy any turn-three game card for card.
Four charts
- Won by our solver
- Undecided
- Proven impossible
- Can’t tell (undecided or impossible)
Chart 1 as a table
| Game | Deals | Won | Undecided | Proven impossible |
|---|---|---|---|---|
| FreeCell (deals 1–32,000) | 32,000 | 99.997% | 0% | 0.003% |
| Golf | 5,000 | 45.08% | 0% | 54.92% |
| Scorpion | 3,000 | 12.47% | 36.17% | 51.37% |
| Yukon | 3,000 | 68.1% | 28.8% | 3.1% |
| Pyramid (3 passes) | 3,000 | 62.83% | 26.9% | 10.27% |
| Canfield (turn 3) | 5,000 | 58.56% | 32.46% | 8.98% |
| Klondike, turn 1 | 3,000 | 68.87% | 31.13% (undecided or impossible) | can’t tell |
| Klondike, turn 3 | 3,000 | 64.27% | 35.73% (undecided or impossible) | can’t tell |
| Spider, 1 suit | 3,000 | 83.37% | 16.63% (undecided or impossible) | can’t tell |
| Spider, 2 suits | 2,000 | 75.45% | 24.55% (undecided or impossible) | can’t tell |
| Spider, 4 suits | 400 | 28.75% | 71.25% (undecided or impossible) | can’t tell |
| Forty Thieves | 600 | 14% | 86% (undecided or impossible) | can’t tell |
- Our solver (dot, 95% interval)
- Published figure (diamond)
Chart 2 as a table
| Game | Ours | Ours, 95% interval | Published | Source |
|---|---|---|---|---|
| Golf | 45.08% | 43.71–46.46% | 45.109% | Blake & Gent, 10 million deals |
| FreeCell, deals 1–32,000 | 99.997% | exact | 99.997% | Keller FreeCell FAQ, via Wikipedia (only #11982 impossible) |
| Klondike, turn 1 (minimum) | 75.63% | 74.07–77.14% | 90.48% | Blake & Gent |
| Klondike, turn 3 (minimum) | 64.27% | 62.53–65.96% | 81.945% | Blake & Gent |
| Canfield (minimum) | 58.56% | 57.19–59.92% | 67.562–71.245% | Blake & Gent (two rule sets) |
- Won
- Proven impossible
- Undecided
Chart 3 as a table
| Bar | Won | Proven impossible | Undecided |
|---|---|---|---|
| All 3,000 deals | 374 | 1,541 | 1,085 |
| 150 undecided deals given a 10× longer search | 1 | 94 | 55 |
- Median
- 10th to 90th percentile
Chart 4 as a table
| Game | Median | 10th percentile | 90th percentile | Wins counted |
|---|---|---|---|---|
| Scorpion | 52 | 50 | 53 | 373 |
| Pyramid (3 passes) | 70 | 59 | 77 | 1,847 |
| Yukon | 94 | 88 | 100 | 1,994 |
| FreeCell (deals 1–32,000) | 101 | 87 | 118 | 31,996 |
| Spider, 1 suit | 112 | 106 | 119 | 2,449 |
| Klondike, turn 3 | 122 | 107 | 141 | 1,879 |
| Klondike, turn 1 | 130 | 119 | 146 | 2,014 |
| Spider, 2 suits | 156 | 140 | 175 | 1,482 |
| Canfield (turn 3) | 171 | 140 | 210 | 2,868 |
| Forty Thieves | 228 | 220 | 236 | 84 |
| Spider, 4 suits | 234 | 201 | 264 | 109 |
How we did it
- The games are ours. Each game on Colouring Nest has a rules engine and a solver in plain JavaScript (the same code that checks every deal our game pages offer can be won). We ran that code, unchanged apart from a counter, on numbered deals: deal 1, deal 2, deal 3 and so on, up to the sample size in the table. FreeCell uses the classic Windows numbering, so our deal 617 is everyone’s deal 617.
- Every win is checked. When the solver says “won”, we replay its moves one by one through the game’s rules and only count the deal if the replay ends in a win.
- “Impossible” means searched to the end. For FreeCell, Golf, Pyramid, Yukon, Scorpion and Canfield, the solver considers every legal move except ones that can never help (for example, moving a lone card from one empty column to another). If it runs out of positions before its budget, no sequence of moves wins the deal. A separate brute-force program with no shortcuts at all re-checked 125 of these proofs (25 per game) and confirmed every one.
- Budgets, not stopwatches. Each deal gets a fixed budget of positions to explore (from 40,000 for one-suit Spider to 300,000 for Golf and Pyramid). Deals still open at the budget got a second look with 5 to 10 times more, where time allowed. Deals that stay open are reported as undecided, never guessed.
- Same deals across modes. Klondike turn one and turn three were run on the same 3,000 layouts; Spider’s suit levels on the same card order recoloured. That makes the comparisons fair.
- Anyone can repeat it. Nothing is random: deal n is always the same layout and budgets are counted in positions, not seconds, so any computer gives the same results. The summary of every game and mode is published as JSON.
The rules we used
Klondike: turn one or three, unlimited passes through the stock. FreeCell: four free cells, nothing back from the foundations. Spider: two packs, one, two or four suits. Pyramid: pairs that make 13, three passes, a win is the pyramid cleared. Golf: one pass, no Ace–King wrap, a Queen may go on a King. Yukon: no stock, any face-up group moves. Scorpion: build in suit, Kings only into spaces, three cards held back. Forty Thieves: two packs, build in suit, one pass. Canfield: turn three, unlimited passes, the top card or the whole column moves. Full rules are under “How to play” on each game’s page.
What this doesn’t tell you
- How often people win. Our solver sees face-down cards and the stock order. A person can’t, so real win rates are much lower than these figures.
- The shortest win. Move counts in the charts are the length of the winning line our solver found, not the minimum.
- Exact odds for Klondike, Spider and Forty Thieves. Our solver’s numbers for these are minimums. For Klondike and four-suit Spider, published research has settled the question far more precisely (four-suit Spider: at least 97.3% of four-suit deals, against our solver’s 28.75%), and we cite it.
- Other rule sets. Small rule changes move these numbers a lot: three passes or one in Pyramid, which cards can move in Canfield, whether any card or only a King can fill a space in Scorpion.
- An independent check of every proof. The 125 proofs we re-checked by brute force were the quickest ones. The others depend on the solvers’ code, which we read line by line; a bug in a game’s rules would make a proof wrong. Wins are safer: each one was replayed.
Cite this study
Colouring Nest (2026). “How winnable is solitaire? The odds of winning 9 games, from 63,000 deals.” https://colouringnest.com/learn/how-winnable-is-solitaire/ (data: https://colouringnest.com/data/solitaire-winnability.json)
The data file holds the results of running Colouring Nest's solitaire solvers on 63,000 numbered deals of 9 games (FreeCell, Klondike turn 1 and 3, Spider 1, 2 and 4 suits, Pyramid, Golf, Yukon, Scorpion, Forty Thieves, Canfield): for each game, how many deals the solver won, proved impossible or left undecided, with 95% intervals and the length of the winning lines.
Quoting a number? Please keep its meaning: “can be won by someone who can see every card”, and for Klondike, Spider and Forty Thieves “our solver won at least”. Questions or a correction: support@colouringnest.com.