How to play Klondike
Twenty-eight cards are dealt into seven tableau piles — one card in the first, seven in the last, only the top card of each face up. The remaining twenty-four form the stock. You win by moving all 52 cards to the four foundations, which build up by suit from the ace: A, 2, 3 … K.
In the tableau, cards build down in alternating colours — a red six goes on a black seven. A face-up run in sequence may be moved together onto another pile. Turning over a face-down card is almost always progress; an empty column takes only a king (or a run starting with one). Click the stock to turn cards into the waste — three at a time in the classic game, one at a time in the easier one — and the top waste card is playable. When the stock runs out, click it again to recycle the waste. Passes here are unlimited.
The rule everyone gets wrong — and what it is worth
Only a king may enter an empty column. Plenty of solitaire software quietly lets any card into a space, and it does not feel like cheating because it is how you wish the game worked. It is worth exactly 13 percentage points: Blake and Gent's Solvitaire study measured draw-3 Klondike at 81.9% winnable with kings-only and 95.0% winnable with any card allowed into a space. (With no moves into empty columns at all the game collapses to 2.2%.) An implementation that relaxes this rule is not slightly generous — it is a different, much easier game wearing the same name.
This board enforces kings-only, and the engine's test suite measures the winnability rate rather than asserting the rule was written down: our solver, run over fresh random deals under the exact published ruleset, reproduces the published rates within its confidence interval — draw 3 against 81.5%, draw 1 against 90.2%.
Winnable-only, proved — not promised
Roughly one draw-3 deal in five cannot be won by anyone. Most sites deal you those anyway and let you assume the loss was your fault. Here, the default modes only deal a board after a solver running in your own browser has found a complete winning line for it — and then replayed that line through the ordinary game rules to check itself. The rail shows the proof: how many moves the line is and how many positions were searched.
Two limits, stated rather than hidden. First, the proof is about thoughtful Klondike — the solver sees every card, including face-down ones, and you do not, so a proved-winnable deal can absolutely still be lost. That is what undo is for, and undo is unlimited. Second, the filter has a lean: a deal the solver fails to crack inside its budget is skipped, and some of the skipped ones were winnable. What is guaranteed is narrower and worth more: it never deals you a board it has not actually won.
A hint that tells you what it is, and a verdict that can say "I don't know"
The Hint button searches the current position. If it finds a full winning line, it says so — the move shown is on a proven winning line, with the count of moves left. If it exhausts the position and finds no win, it tells you the position is dead rather than politely suggesting moves that cannot matter. And if it runs out of budget, it says the hint is a guess, not a proof — which is the label every other site's hint deserves and never carries.
The Still winnable? button asks the full question of the whole position and gives one of three answers: still winnable (a line exists, found and counted), proven lost (every reachable position searched, no win among them), or don't know (the search hit its budget). The third answer is not a soft version of the second, and this page never presents it as one.
Scoring, and why undo reverses it
Standard scoring, the one Windows made familiar: 5 for waste to tableau, 10 for any card reaching a foundation, 5 for turning a face-down card, −15 for taking a card back off a foundation, and a recycle penalty for extra passes through the stock (−20 after the third pass at draw 3, −100 after the first at draw 1; never below zero). Undo takes the points back too — the score describes the position you are in, not how many attempts it took to get there.
Strategy that actually helps
Turn face-down cards over before anything else. Every hidden card is information you are playing without. Given a choice of plays, prefer the one that flips a card, and prefer flipping in the tall piles on the right — they hide the most.
Do not empty a column without a king ready. An empty column that sits empty is a wasted move; only a king can use it, so dig out a buried king first and know which colour you want — the king's colour dictates the whole alternating run you will build on it.
Aces and twos go up immediately; after that, send cards home carefully. A card on a foundation can still be taken back here (for −15), but the better habit is not to strand it in the first place: a black five sent home cannot hold the red four you find two moves later.
In draw 3, count in threes. Only every third stock card surfaces, so taking one card from the waste shifts which cards you will ever see again. If the stock has cards you need at the wrong depth, look for a waste card you can play purely to change the phase.
Questions people ask
Is every Klondike deal winnable?
No. With every card's identity known, 81.5% of draw-3 deals and 90.2% of draw-1 deals are winnable (Blake & Gent, without cards coming back off foundations) — so about one draw-3 deal in five is lost before you touch a card. Our solver reproduces both rates in the test suite. The winnable-only modes exist so that number stops applying to you.
Can any card go into an empty column?
Only a king, under the real rules. Software that allows any card is playing a game that is 13 percentage points easier — 95.0% winnable instead of 81.9% — and that difference is measured, not folklore.
Draw 1 or draw 3 — which should I play?
Draw 3 is the classic; draw 1 is genuinely easier (90.2% vs 81.5% winnable) because every stock card is reachable on every pass, so you cannot lock a card away by mistiming the talon. Both are here, both with unlimited passes.
What does "proved winnable" mean exactly?
A solver in your browser found a complete winning line for this exact deal and replayed it through the ordinary rules before you saw a single card. It proves the deal is winnable with perfect information; it does not promise you will win it, since you cannot see the face-down cards. And the filter skips deals it cannot solve in budget, some of which were winnable — it never claims otherwise.
Are your deal numbers the same as Windows Solitaire?
No. Windows Solitaire never published a deal numbering (FreeCell did — and our FreeCell uses the real Microsoft numbers). Klondike deal numbers here are our own: the number fully determines the deal, so the same number is the same board forever, and every deal has a permalink. They just will not mean anything on another site.
Is this solitaire free?
Yes. No signup, no ads, no video between games, no daily limit, no premium tier, no paid hints, and no ad-free subscription — because there are no ads to remove.
The honest small print
The solver is a depth-first search with a transposition table, compound talon moves and safe autoplay — a research-grade technique, not a toy, but bounded by a node budget. It has three answers and the page keeps them apart: winnable (a line was found and replayed through the rules before being trusted), proven lost (every reachable position exhausted), and could not decide (budget ran out — a fact about the search, never a claim about the deal). Published winnability rates are for thoughtful Klondike — perfect information — and our suite reproduces them by sampling the same distribution with our own solver, which is a statistical agreement with an error bar, not a replication of the original instances. Win counts and best times are stored unverified, because a browser can lie about its own score and we would rather label that than pretend otherwise.