Klondike patience, checked against the science
Klondike is the patience game many people picture when they hear the word "solitaire": seven columns, a stock pile in the corner, and four Aces waiting to be built up to Kings. This page deals it the gentler way, turning one card from the stock at a time, and you can go round the stock as often as you like.
We wanted to know two things about this game. First, how many deals can be won at all? Second, how many do players actually win? The first question has been studied properly. In February 2026 two computer scientists from the University of St Andrews in Scotland, Charlie Blake and Ian Gent, published the most precise answer yet in the Journal of Artificial Intelligence Research. Rather than simply repeat their figure, we built our own version of this site's Klondike in Python, dealt thousands of numbered games, and set our results beside theirs. Where we agree, you can trust the number twice over. Where we differ, we explain why.
The short version: with every card visible, about nine deals in ten can be won at Turn 1. Our own computer player, which sees only what you see and never takes a move back, won 42.3% of 10,000 deals.
Our figures next to the published study
| What was measured | Our study (September 2026) | Blake & Gent, St Andrews (2026) | Do they agree? |
|---|---|---|---|
| Deals winnable with every card visible (Turn 1) | At least 76.6% proven; 2.1% proven impossible; 21.3% undecided (1,000 deals) | 90.480% (±0.116%), from 1,000,000 deals | Yes: 90.48% sits inside our range of 76.6–97.9% |
| Deals that cannot be won at all (Turn 1) | Between 2.1% and 23.4% | 94,629 of 1,000,000, about 9.5% | Yes, within our range |
| Deals won by a player who cannot see hidden cards and never undoes | 42.3% (95% range 41.3–43.3%, 10,000 deals) | Not studied | No published comparison |
| Real players' win rate on a large solitaire site | 44.8% of 913,106 Klondike games in the month before 16 July 2025, reported by solitairex.io | A different population and different rules, so a comparison, not a check | |
The pattern is worth pausing on. A computer that could see every face-down card would win roughly nine games in ten. A careful player who cannot see them wins far fewer, and our simple rule-following player landed at 42.3%, close to the 44.8% that real people managed on solitairex.io. The gap between "could be won" and "was won" is almost entirely about hidden information and choices, not about the deal itself.
How Klondike works on solitaire-free.uk
Setting out the cards
Twenty-eight cards are dealt into seven columns: one card in the first, two in the second, and so on up to seven. Each column shows just its lowest card face up. The other 24 cards form the stock.
Building
- In the columns, build downwards in alternating colours: a black 9 on a red 10, a red 4 on a black 5.
- You may move any face-up run that is already in order, not just the last card.
- When a face-down card is uncovered, it turns over.
- Only a King, or a run headed by a King, may be placed in an empty column.
- On the foundations, build each suit upwards from Ace to King.
The stock
Click the stock to turn one card onto the waste pile. Its top card can be played to a column or a foundation. When the stock runs out, turn the waste back over and go again, as many times as you need. You may also bring a card back down from a foundation if you need it to catch a lower card.
What the comparison teaches you at the table
Most deals are not the problem
If roughly nine deals in ten are winnable with perfect information, a lost game is far more often a missed line than a hopeless deal. That is the most useful thing the St Andrews figure tells an ordinary player. Before you blame the cards, use Undo and try a different order.
Hidden Aces are common, but they rarely decide the game
In 88% of our 10,000 deals, at least one Ace began face down in the columns, with 1.6 of the four hidden on average. It sounds grim, yet our player won 41.6% of deals with two or more Aces hidden, against 44.5% when none were hidden. Digging out buried cards matters, but a buried Ace is not a lost game.
Turn cards over before tidying
Our player's second priority, after sending cards to the foundations, is any move that uncovers a face-down card, starting with the column that hides the most. The published winnability figure assumes perfect knowledge; you get closer to it by turning over as many cards as early as possible, because every card you can see is information the solver had and you did not.
Expect to go through the stock more than once
Games our player won used 2.3 passes through the stock on average. Going round again is normal, not a sign that the game is slipping away.
A dead start happens
In 8.6% of our deals there was no move at all among the seven face-up cards. That is when you go straight to the stock. It is not a reason to deal again.
Playing on solitaire-free.uk
Klondike on solitaire-free.uk is free, with nothing to download, no account and no email address. Any up-to-date browser will do, whether on a computer, a tablet or a phone.
- Large, clear cards and a zoom control for comfortable reading.
- Hint, unlimited Undo and Restart.
- Auto Play to send the cards home once the game is effectively won.
- Move cards by dragging them, or by tapping the card and then its destination.
- Wins, time and moves shown for the current game; sound, fullscreen and table colour can all be changed.
Prefer a stiffer test? Solitaire Turn 3 uses the same layout with a much harder draw, and FreeCell puts every card on show from the start.
How we ran the research, step by step
The solitaire-free.uk team ran this study in September 2026. Our aim was not to re-derive a famous number for its own sake, but to measure this site's version of Klondike and see whether it behaves as the published research says it should.
- Reading the rules from our own game code. We took the rules from the code that runs this page: unlimited passes through the stock, runs may move as a group, only Kings fill an empty column, and one card may come back down from a foundation.
- Rebuilding the game as a program. We wrote Klondike again in Python so a computer could play it. The computer player's moves were checked against those rules over 2,000 games.
- Dealing numbered games. Each deal is shuffled from a numbered seed (deals 1 to 10,000), so exactly the same games can be dealt again. These are our own numbers, not the deal numbers you see on the site.
- Choosing two kinds of player. A rule-following bot that sees only face-up cards and never undoes a move gives a floor: a patient person should do better. A solver that is allowed to see every hidden card gives a ceiling: nobody can do better. The solver searched up to 150,000 positions per deal.
- Running the games. The bot played all 10,000 deals. The solver, which is much slower, worked through 1,000.
- Comparing with the published study. Blake and Gent report 90.480% for this variant with every card known. Our solver proved 76.6% winnable and 2.1% impossible, leaving 21.3% undecided because it ran out of search budget. The published figure falls inside our range, which is what should happen if both pieces of work are sound.
- Stating the margin of error. A "95% range" is the band the true figure very probably sits in, given how many deals we played. For the bot, 42.3% means somewhere between 41.3% and 43.3%.
- Being honest about the limits. Our bot is a simple player, so its 42.3% understates what a thoughtful human with Undo can achieve. Our solver's undecided deals are exactly that: undecided, not lost.
Behind the numbers: the code
We wrote the study in Python 3.11, relying on nothing but its standard library. Klondike lives in one module, klondike.py, alongside a shared runner that spread the deals across 19 worker processes on a 20-core machine. The bot follows a fixed list of priorities: foundation moves first, then moves that uncover hidden cards, then Kings into empty columns, then waste-to-column plays, and it draws only when nothing else helps. The solver is a depth-first search that tries every legal move, including moving part of a run and taking a card back from a foundation, and keeps a record of positions it has already seen. Its only shortcuts are moves that are provably safe, so when it runs out of options that counts as proof that the deal cannot be won. Ranges for percentages use the Wilson score interval. The bot's 10,000 games took well under a minute; the solver's 1,000 deals took about seventeen minutes.
We used an AI assistant, Claude, made by Anthropic, to help write and review the Python and to draft this page. None of the numbers came from the AI: every figure is the output of running the code, and we checked it against the St Andrews study before publishing.
Frequently asked questions
Who worked out that about 90% of Klondike deals can be won?
Charlie Blake and Ian Gent of the University of St Andrews, in a paper published in the Journal of Artificial Intelligence Research in 2026. Their Turn 1 figure is 90.480%, assuming the player knows where every card is.
Why is your proven figure lower than theirs?
Because we gave our solver a far smaller search budget and far fewer deals. It proved 76.6% winnable but could not decide 21.3%. Their figure lies inside our range; ours is simply less precise.
Does anybody really win 90% of games?
Not without seeing the face-down cards. Real players on solitairex.io won 44.8% of 913,106 games, and our no-peeking bot won 42.3%. The 90% figure is a ceiling, not a target.
Is going through the stock many times cheating?
No. This version allows unlimited passes by design. Our winning games used about 2.3 passes on average.
Where can I read more about the other games?
Every game has its own page with the same side-by-side comparison. Try Golf, where one small rule doubles your chances, or browse the full list on the games page. Longer write-ups appear on the blog.
By Admin, solitaire-free.uk. Study date: 18 September 2026.