Klondike Solitaire: how many deals can actually be won?
Every Klondike player eventually asks the same thing after a loss: was that deal ever winnable, or did I blow it? We wanted a real answer for the version on this page, so we handed 1,000 shuffled deals to a computer solver that is allowed to see every card, including the ones lying face down. Nothing a person does can beat that kind of knowledge, which makes its results a ceiling for what is possible.
The short answer: at least 76.6% of Turn 1 deals can be won with perfect play, and only 2.1% are provably impossible. The remaining 21.3% were too complex for the solver to settle within its search limit, so the true share of winnable deals lies somewhere between 76.6% and 97.9%. Put simply, if you lose a game of Turn 1 Klondike here, the odds are good the deal had a way through.
This is the classic game most Americans picture when they hear "solitaire": seven columns, a stock pile, four foundations to fill from Ace to King. It is free to play right here in your browser, with nothing to download and no account to create.
The ceiling at a glance
| What we measured | Result | Deals | How |
|---|---|---|---|
| Deals solved with perfect play | 76.6% (95% range 73.9–79.1%) | 1,000 | Solver that sees every card |
| Deals proven impossible | 2.1% (95% range 1.4–3.2%) | 1,000 | Same solver searched every line |
| Deals left undecided | 21.3% | 1,000 | Search limit of 150,000 positions reached |
| Winnable share, therefore | between 76.6% and 97.9% | 1,000 | Solved, plus solved-or-undecided |
| Published estimate, same idea | 90.48% (±0.12%) | about 1,000,000 | Blake & Gent, 2026 |
| What a simple no-peeking player wins | 42.3% (95% range 41.3–43.3%) | 10,000 | Rule-based bot, never undoes |
The published figure from the Blake & Gent study in the Journal of Artificial Intelligence Research sits comfortably inside our range. Their solver ran far longer on far more deals, so their number is the sharper one; ours shows the same picture from an independent program built on this site's exact rules.
Why "winnable" and "won" are different numbers
There is a wide gap between a deal that can be won and a deal a person does win. Our perfect-knowledge solver clears at least three deals in four. A simple computer player that sees only the face-up cards, just as you do, and never takes a move back, wins 42.3%. Real people land close to that: on solitairex.io, players won 44.8% of 913,106 Klondike games in the month before 16 July 2025.
That gap is the whole opportunity. The cards are rarely the problem; hidden information and early commitments are. Each time you uncover a face-down card, you learn something the perfect solver already knew. The better you play around what you cannot see yet, the closer you climb toward that ceiling.
The rules as this site plays them
How the cards start
One standard 52-card deck. Seven columns hold 1 to 7 cards, with only the top card of each turned up. The other 24 cards form the stock in the top-left corner. The four foundations, top right, start empty.
Where cards can go
- In the columns, build downward in alternating colors: a black 9 on a red 10, a red 4 on a black 5.
- Any face-up run that is already in order can be moved as a unit.
- An empty column accepts a King, alone or leading a run, and nothing else.
- Foundations build upward by suit from Ace to King. You may bring one card back down from a foundation if you need it.
Drawing from the stock
This page uses Turn 1: each click turns one card from the stock. When the stock runs out, you can flip the waste pile over and go through again, as many times as you like.
The finish
Move all 52 cards to the foundations. Once everything is face up and in order, Auto Play will finish the job for you.
What the solver teaches about playing well
Most losses are recoverable
With every card visible, at least three deals in four can be won, so a stuck position is more often the result of an earlier choice than a doomed shuffle. Unlimited Undo lets you go back and look for it. Step back to the last point where you had options and try the other branch.
Turn over hidden cards first
The solver's advantage is that it knows what is underneath. You can shrink that advantage by exposing face-down cards early. When two moves look equal, prefer the one that flips a card, and prefer the column with the most cards still hidden.
Treat empty columns as King parking only
On this site only Kings may fill an empty column. Emptying a column with no King ready to move in gains you nothing and can cost you a place to put cards. Clear a column when a King is waiting, not before.
Hold back from the foundations when it matters
Sending a card home is usually right, but a red 6 on the foundation cannot receive a black 5 from a column. If lower cards of the opposite color are still out, think twice.
Use the stock as a resource, not a lottery
With Turn 1 and unlimited passes, every stock card is reachable eventually. There is no need to grab a waste card the moment it appears; if it does not help a column now, leave it and come back.
Inside the test, one step at a time
The solitaire-online.us team ran this study in September 2026 to find the ceiling for Klondike on this site specifically, not for the game in general.
- Taking the rules from this site, not a rulebook. We took the rules straight from this site's game files: unlimited passes through the stock, Kings only into empty columns, any ordered face-up run can move, and one card may come back down from a foundation.
- Rebuilding the game as a program. We wrote a copy of Klondike with those rules. The rule-based player's moves were checked against the rules over 2,000 games (1,000 at Turn 1 and 1,000 at Turn 3).
- Dealing the cards. Deals 1 to 1,000 for the solver and 1 to 10,000 for the rule-based player. Each deal comes from a numbered seed, so the same deals can be produced again. These are our own deal numbers, not the site's.
- Choosing the players. A "thoughtful" solver that sees every card and tries every legal move gives the ceiling. A rule-based player that sees only face-up cards and never undoes gives a floor.
- Running it. The solver had up to 150,000 positions per deal. If it found a route to victory, the deal counted as solved. If it ran out of positions to try without the limit being reached, the deal was proven impossible. If the limit came first, the deal stayed undecided.
- Checking against published research. Blake & Gent (2026) report 90.48% for Turn 1 with the same "see every card" convention. That value falls inside our range of 76.6–97.9%.
- Working out the margin of error. A 95% range means that if we repeated the test on fresh deals, the true value would land inside that range 19 times out of 20. For deals solved, that is 73.9–79.1%.
- What the numbers cannot tell you. A solver that sees the face-down cards is not how anyone plays. Its number is what is possible, not what is likely. And about one deal in five stayed undecided, which is why we quote a range.
The code behind the ceiling
The study is written in Python 3.11 using only the standard library. Klondike lives in its own module, klondike.py, which holds both the solver and the rule-based player; a shared runner, runner.py, spreads deals across 19 worker processes on a 20-core Windows machine. Each deal is shuffled with Python's random number generator from a seed like "klondike:42", which gives the same shuffle every time.
The solver is a depth-first search: it tries a move, then the next, and backs up when it hits a dead end. It keeps a set of positions it has already visited so it never searches the same one twice, and it only makes automatic foundation moves that are provably safe, so an exhausted search is a genuine proof. The 1,000-deal solver run took about 17 minutes; the 10,000-deal rule-based run took about 14 seconds. The 95% ranges use the Wilson score interval, a standard method for percentages.
We used an AI assistant, Claude by Anthropic, to help write and review the simulation code and to draft this page. None of the figures came from the AI: every number is output from running the code, and our team compared the results with the published research before putting them here.
Because each shuffle comes from a numbered seed, anyone with the same code gets exactly the same deals and the same verdicts.
By Admin, solitaire-online.us. Study date: September 18, 2026.
Ceiling questions from players
Is every game of Klondike solitaire winnable?
No. In our test 2.1% of Turn 1 deals were proven impossible even with every card visible, and 21.3% were too complex to settle. The published study puts Turn 1 at 90.48% winnable, so roughly one deal in ten cannot be won however well you play.
Why do I win far less often than the solver?
Because the solver can see face-down cards and you cannot. Real players on solitairex.io won 44.8% of their games, very close to our no-peeking player's 42.3%.
Does Turn 3 change the odds?
A lot. With every card visible, our solver proved 9.5% of Turn 3 deals impossible, compared with 2.1% at Turn 1. See the Turn 3 page for the full picture.
Does Undo count against me?
Not here. There is no leaderboard to protect, and Undo is the closest a person can get to the solver's habit of trying a line and backing out.
Does it work on a phone or tablet?
Yes. It runs in any modern browser on phone, tablet or computer. Tap a card and then where it should go, or drag it, and use the zoom control for bigger cards.
Which game should follow Klondike?
If you want a game where almost every deal is winnable, try FreeCell. For a tougher ceiling, look at Baker's Solitaire. Every other game is listed on the games page; more findings from the study are on our blog.