Golf Solitaire: an exact answer, and it's 92.7%

Golf is a race to clear the table. Seven columns of five cards lie face up across the table, 35 in all. You move cards one at a time from the bottom of any column onto a single discard pile, as long as each one is one rank above or below the card on top. When you run out of plays, you turn a card from the stock and carry on.

Golf is small enough for a computer to check every possible line of play, so for once we can give an exact answer rather than a range. We ran every one of 10,000 deals to the end. With the rules used on this site, 92.7% of deals can be cleared with perfect play. The other 7.3% cannot, whatever you do.

There is a catch that matters. That perfect play assumes you know the order of the stock, which you don't. Even so, the ceiling tells you something useful: almost every Golf deal here has a route to victory. You can play it for free in any browser.

The numbers behind the 92.7%

The published classic-rule figure is from Blake & Gent (2026), Journal of Artificial Intelligence Research. Our classic-rule result, 44.0%, sits close to it, which is a good sign that our solver is correct.

The wrap doubles the ceiling

On this site, Golf lets you play a King on an Ace and an Ace on a King. In classic Golf the sequence stops at the ends. That one rule is the difference between a game where fewer than half of deals can be cleared and one where more than nine in ten can.

The reason is simple. Without the wrap, a King on the discard pile is a dead end: only a Queen can follow. An Ace is the same, with only a 2. With the wrap, every card has two neighbors, so a run can keep going around the corner. It is one of the largest effects of any single rule we measured.

Ceiling versus reality

Here is the sobering part. A greedy computer player that cannot see the stock, and always takes the tableau card that seems to lead to the longest run, cleared just 22.4% of deals with the wrap. That is a quarter of the ceiling. Golf looks simple, but the gap suggests that many winnable deals need you to pass up an obvious play now to keep a better one for later.

The solver's solutions typically left about one stock card unused, so a winning game often comes right down to the wire.

Golf rules on this site

• One 52-card deck. Seven columns of five cards each, all face up.
• One card starts the discard pile; the remaining 16 form the stock.
• Play the bottom card of any column onto the discard pile if it is one rank higher or lower, in any suit.
• Kings and Aces wrap: a King can go on an Ace and an Ace on a King.
• When you have no play, turn the next stock card onto the discard pile. There is no redeal.
• Clear all 35 table cards to win.

How to play closer to 92.7%

Trace every chain before you commit

Before each play, trace how far each option could run. A card that leads to a chain of four is usually better than one that leads to a chain of one.

Save your "turning" cards

Kings and Aces are special here because they let a run turn around. Try not to spend them when an ordinary card would do.

Don't clear one column at the expense of the rest

Evenly shortening columns keeps more cards available. A single tall column at the end is often what sinks a game.

Expect a slow start sometimes

In about 27% of deals, the first discard card has no play at all in the columns. That is normal; turn a stock card and carry on.

Count what's left in the stock

Solutions usually finish with around one stock card to spare. If you are down to three stock cards with ten table cards left, look hard for a longer chain.

Golf on the ceiling ladder

Golf's 92.7% puts it near the top of every game we measured. Only FreeCell settled higher. For context:

• FreeCell: 99.68% solved, no deal proven impossible.
• Golf (this site): 92.7% winnable, exact.
• Pyramid: 62.2–86.0% winnable with three passes.
• Baker's Solitaire: 31.8% winnable, exact.

Golf and Baker's are the only two games where our solver settled every deal, so their numbers are the most solid in the study. They sit at opposite ends of the ladder, and in both cases the site's own rule is the reason: the King-and-Ace wrap lifts Golf, and the Kings-only empty-column rule holds Baker's down.

Inside the test, one step at a time

The solitaire-online.us team ran this study in September 2026 to get an exact ceiling for Golf on this site.

1. Reading the rules from our game code. Seven columns of five, a 16-card stock with no redeal, plus-or-minus-one building in any suit, and the King-and-Ace wrap, which is the only Golf variant this site offers.
2. Rebuilding the game. We coded Golf with those rules, and a second version without the wrap for comparison.
3. Dealing the cards. We played deals numbered 1 to 10,000 under the site rules, and deals 1 to 3,000 under the classic no-wrap rules.
4. Choosing the players. An exhaustive solver that knows the stock order and checks every line, giving an exact yes or no for each deal. A greedy player that doesn't know the stock, for comparison.
5. Running it. Every deal was searched to the end; none was left undecided.
6. Checking against published research. Without the wrap we found 44.0%, close to Blake & Gent's published 45.109% for classic Golf. No published figure covers the wrap rule.
7. Working out the margin of error. Even exact results for 10,000 deals are a sample of all possible shuffles. In 19 of 20 repeats, the true share would lie in our 95% range of 92.2–93.2%.
8. What the numbers cannot tell you. The solver knows the stock order. A real player does not, so the practical ceiling is lower. How much lower depends on skill.

The code behind the ceiling

We wrote everything in Python 3.11 with the standard library only. Golf lives in golf.py, and the shared runner.py spread the deals across 19 worker processes on a 20-core machine. Each deal is regenerated from its seed.

The solver is an exhaustive depth-first search over which cards remain in each column, how far through the stock you are and the rank on top of the discard pile. It remembers every dead-end position so it never explores the same one twice, which lets it finish every deal with a firm answer. The greedy player instead picks the table card with the longest visible follow-up and turns a stock card only when stuck. The run with the wrap took about 12 minutes; without it, about 7. Ranges use the Wilson score interval.

An AI assistant, Claude from Anthropic, helped write and review the code and draft this page. It did not produce the numbers. Every figure is output from the solver, and we checked the no-wrap result against the published figure before trusting the rest.

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

Can every Golf Solitaire game be won?

No. With this site's rules, 92.7% of our 10,000 deals can be cleared with perfect play; the other 7.3% cannot.

Why do I win far less often than 92.7%?

Because perfect play means knowing the stock order in advance. A simple player that doesn't know it cleared only 22.4%.

Does Golf here let a King follow an Ace?

Yes. This site's Golf wraps around, so King-on-Ace and Ace-on-King are both legal. That rule roughly doubles how many deals can be won.

Is Golf a good quick game?

Yes. With only 35 table cards and 16 in the stock, a game is over quickly. For another single-deck puzzle with a different twist, try Pyramid.

What has an even higher ceiling?

FreeCell, where our solver proved no deal impossible. The full list is on the games page.