Spider Solitaire, four suits: nearly every deal can be won, almost nobody wins them

Four-suit Spider is the full game: two decks, all four suits, 104 cards in ten columns. It has a fearsome reputation, and on the question this site cares about, how many deals are winnable, it holds a surprise.

The most thorough published study, by Blake and Gent at the University of St Andrews, estimates that about 98.5% of four-suit Spider deals can be won by a player who can see every card. By that measure the ceiling is almost as high as FreeCell's. Yet our simple computer player, which sees only face-up cards and never undoes, won just 4 of 2,000 deals: 0.2%.

No other game on this site shows a bigger gap between what is possible and what a straightforward player achieves. That gap is the skill in four-suit Spider. You can play it here for free, in any browser, with nothing to download.

The widest gap we found

The ceiling is from Blake & Gent (2026), Journal of Artificial Intelligence Research; their Spider figure has a wider margin than most in the paper because Spider is so hard to solve by computer. The player tier comes from the solitairex.io homepage, which does not say how it was measured.

How the ladder climbs from one suit to four

With four suits, our player rarely finished even one suit in a game. Runs can only be moved as a group when they are all one suit, and with four suits in play, nearly every run it built was mixed. Every mixed run is a block that has to be dismantled card by card. See one suit and two suits for the gentler rungs.

The game on this site

• 104 cards, all four suits, in ten columns: four of six cards and six of five. Top card face up.
• 50 cards in the stock, dealt one to every column per click.
• Any card goes on a card one rank higher; only same-suit runs move as a group.
• Any card may go into an empty column.
• As soon as one suit runs from King down to Ace, the game lifts it off the table.
• Clicking the stock with a column empty brings up a warning.

Reaching toward that 98.5%

Sort before you build

A long mixed run feels like progress but moves like a brick. Early on, the priority is to uncover cards and keep what you build in suit.

Keep an empty column as long as you can

An empty column is the only place to unpick a mixed run. Our player managed one before its first deal in just 6 games and won 2 of them; that is a small sample, but it hints at how valuable the space is.

Think in suits, not ranks

The move that makes the tidiest single-suit run is usually right, even if a different move looks longer.

Deal only when the table is in the best order you can manage

Each deal drops a card on every column. Make sure no available move would uncover a hidden card first.

Use Undo as your map of hidden cards

The published ceiling assumes perfect knowledge. Undo is how a person learns what is under a card and tries again, and in four-suit Spider it is the difference between 0.2% and a real chance.

Four-suit Spider on the ceiling ladder

On the published ceiling alone, four-suit Spider ranks near the top of the games on this site:

• FreeCell: 99.68% solved in our test; 99.998881% published.
• Spider, four suits: 98.487% published (±1.513%).
• Golf: 92.7% in our test.
• Klondike Turn 1: 90.48% published.

Yet on the floor, measured by our simple players, four-suit Spider ranks dead last. No other game on this site shows a wider gap between what is possible and what a simple strategy achieves, which makes four-suit Spider the clearest case of a game decided by skill.

Inside the test, one step at a time

The solitaire-online.us team ran this part of the study in September 2026. For four-suit Spider we measured the floor ourselves and rely on published research for the ceiling.

1. Reading the rules from our game code. The layout, the 50-card stock dealt ten at a time, same-suit group moves, empty-column rule and automatic removal of finished runs come from this site's game files. We assumed that dealing onto an empty column is not allowed; the game warns you, but we did not confirm it in a browser.
2. Rebuilding the game. The same Spider program as the one- and two-suit tests, with all four suits.
3. Dealing the cards. Seeds 100,001 to 102,000. The player was tuned on seeds 1 to 30, which were not used in the results.
4. Choosing the player. A look-ahead player that searches 250 positions per turn using only the visible cards and never undoes. We did not build a four-suit Spider solver, so the ceiling comes from the published study.
5. Running it. 2,000 deals, each played until won or until no move improved the position.
6. Checking against published research. Blake & Gent (2026) found 98.487% (±1.513%) winnable with every card visible. We cannot check that ourselves; we report it as they give it.
7. Working out the margin of error. With only 4 wins, our range is wide in relative terms: 0.08% to 0.51%. In 19 repeats out of 20, the true rate for this player would fall inside it.
8. What the numbers cannot tell you. Our player is very weak at four suits. Its 0.2% says nothing about how winnable the deals are; it shows how little a simple strategy gets you.

The code behind the ceiling

The simulation is Python 3.11 with only the standard library. The Spider code is in spider.py, and runner.py divides the 2,000 deals between 19 worker processes on a 20-core machine. Every deal comes from a numbered seed.

The player is a best-first search with a budget of 250 positions per turn. It scores positions by adding points for same-suit links, empty columns and newly revealed cards, a few points for mixed links, and subtracting points for each face-down card. It plays toward the best position found and deals when nothing better is available. All 2,000 four-suit games finished in roughly three and a half minutes. Because we built no solver, there is no ceiling run to report. The 95% range uses the Wilson score interval, which behaves well even with very few wins.

We used Claude, an AI assistant from Anthropic, to help write and review the code and to draft this page. The AI generated none of the figures. Ours come from running the code; the ceiling comes from the published paper linked above.

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

How can a game be 98.5% winnable and so rarely won?

By the published estimate, about 98.5% of deals are winnable for a player who can see every card. For a person, the true share you can reach depends heavily on skill and Undo.

Why did your computer player win almost nothing?

It never undoes, cannot see face-down cards and follows a simple scoring rule. With four suits, that leads to mixed runs it cannot move.

What do real players manage?

solitairex.io places four-suit Spider in its hardest tier, with win rates of 11% or less, though it does not publish its method.

Is this the hardest game on the site?

For a simple player, yes. By ceiling, Baker's Solitaire is lower: on this site's rules only 31.8% of its deals can be won at all.

Where should I start instead?

Start with one-suit Spider, then two suits. All the other games are on the games page.