Solitaire Turn 3: the ceiling drops, and some deals simply close

Turn 3 is Klondike with one change: the stock hands you three cards at a time, and only the top one of the three is playable. It sounds like a small tweak. For the question of what can be won at all, it is not small.

We gave 1,000 shuffled deals to a solver that sees every card, including the face-down ones and the exact order of the stock. It found a winning route in 66.2% of them. It proved 9.5% impossible, meaning no sequence of moves, however clever, clears the board. The other 24.3% were too complex to settle within the search limit. So the real share of winnable Turn 3 deals sits between 66.2% and 90.5%.

Compare that with Turn 1 on our home page, where only 2.1% of deals were proven impossible. In our test, drawing three at once more than quadrupled the share of deals proven closed before the first card moves. You can play Turn 3 here for free, in any browser, with no sign-up.

Turn 3 against Turn 1: the ceilings compared

The published values come from Blake & Gent's 2026 study in the Journal of Artificial Intelligence Research, which ran millions of deals. Both of their figures fall inside our ranges.

Why three cards close so many doors

With Turn 1, every card in the stock eventually reaches the top of the waste, because you can cycle through as often as you like. With Turn 3, that is no longer true. The stock is dealt in fixed groups of three. A card sitting second or third in its group only becomes playable once something above it has been used, which shifts where the groups split on the next pass.

So some stock cards are locked in place. If the card you need is buried in the middle of a group, and nothing ever frees the card on top of it, it never comes up. The solver can see this in advance, which is how it proved almost one deal in ten dead on arrival. You cannot see it, which is why Turn 3 feels so much harder in practice: our rule-based player's win rate fell from 42.3% at Turn 1 to 10.6% here.

Rules on this page

• One 52-card deck. Seven columns of 1 to 7 cards, top card face up; 24 cards in the stock.
• Each click on the stock turns three cards to the waste. Only the top waste card can be played.
• When the stock is empty, flip the waste back and go again. There is no limit on passes.
• Columns build down in alternating colors. Ordered face-up runs may move together.
• Only a King, or a King-led run, may fill an empty column.
• Foundations build up by suit, Ace to King. One card may come back down if you need it.

Playing closer to the ceiling

Map the groups of three first

The first time through, you are gathering information. Note which cards show up on top of each group of three. Those are the cards you can count on; the others depend on what you use.

Use a waste card to reshuffle the groups

Playing even one card from the waste shifts every later group of three by one position on the next pass. When a key card is stuck second in a group, taking the card above it, if it fits somewhere harmless, can bring it up next time.

Don't burn waste cards you'll need later

The flip side: playing a waste card just because it fits can also move a card you were counting on out of reach. Before you play from the waste, think about what it does to the next pass.

Uncover columns before relying on the stock

Face-down tableau cards do not have the Turn 3 problem. Every one of them becomes available the moment the cards above it move. If you have a choice, dig into the columns first.

Test waste plays with Undo

The solver wins by seeing the future. Undo lets you peek at it by trying a line and stepping back. On Turn 3, where one waste play changes every later group, that is especially valuable.

Inside the test, one step at a time

The solitaire-online.us team carried out this study in September 2026 to measure how many Turn 3 deals on this site can be won at all, and how many cannot.

1. Reading the rules from our game code. The draw count of three, unlimited passes, Kings-only empty columns and the one-card return from foundations all came from the game files this page runs.
2. Rebuilding the game. Our program models exactly which stock cards can be reached on each pass of three, not an approximation. The rule-based player's moves were checked against the rules over 1,000 Turn 3 games.
3. Dealing the cards. Seeds 1 to 1,000 for the solver and 1 to 10,000 for the rule-based player, the same numbered deals we used for Turn 1, so the two draw modes face identical shuffles.
4. Choosing the players. The solver sees every card, which sets the ceiling. The rule-based player sees only face-up cards and never undoes, which sets a floor.
5. Running it. Up to 150,000 positions per deal for the solver. A deal was solved if it found a win, proven impossible if every reachable position was explored without one, and undecided if the limit came first.
6. Checking against published research. Blake & Gent (2026) put Turn 3 at 81.945% with every card visible. That lies inside our 66.2–90.5% range.
7. Working out the margin of error. A 95% range is the band that would contain the true value in 19 repeats out of 20. For deals proven impossible, ours is 7.8–11.5%.
8. What the numbers cannot tell you. The solver's advantage is complete knowledge of the stock order, which no player has. About a quarter of deals stayed undecided, so we quote a range rather than one number.

The code behind the ceiling

Everything was written in Python 3.11 with nothing beyond the standard library. The game, the solver and the rule-based player live in klondike.py; runner.py farms the deals out to 19 worker processes on a 20-core machine. Each shuffle comes from a numbered seed, so it can be recreated exactly.

The solver is a depth-first search with a record of every position it has already seen. For Turn 3 it tracks the precise state of the stock and waste, so it knows which cards each pass can actually bring to the top. It applies only foundation moves that are provably safe, so an exhausted search counts as proof that a deal cannot be won. The solver run over 1,000 deals took about 17 minutes; the rule-based player got through 10,000 deals in about 8 seconds. For the percentage ranges we used the Wilson score interval.

An AI assistant, Claude from Anthropic, helped us write and check the simulation code and draft this page. It did not supply any results. Every figure is the output of the code itself, and we checked the results against the published study before using them.

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 many Turn 3 games are impossible?

In our 1,000-deal test, at least 9.5% were proven impossible even for a solver that sees every card. The published study puts Turn 3 at 81.945% winnable, which means roughly 18% cannot be won, so many of our undecided deals are probably closed too.

Is Turn 3 just harder, or actually less winnable?

Both. More deals are closed from the start, and the ones that remain open are harder to find a route through without seeing the stock.

Should beginners start with Turn 1?

Yes. At Turn 1, every stock card is reachable eventually, which makes the game far more forgiving. Try the Turn 1 version first.

Why do some cards never seem to come up?

Because in Turn 3 the stock is dealt in fixed groups of three. A card in the middle of a group stays hidden until you play the card above it.

Can I switch draw modes on this site?

This page plays Turn 3; the home page plays Turn 1. For a game where almost nothing is left to chance, try FreeCell, or browse every option on the games page.