What FreeCell Solvers Found: Deal 11982 and the Limits of Search
For about fifteen years, a slow-motion collective effort took place across early internet forums: people were trying to beat all 32,000 numbered FreeCell deals that shipped with Windows. Not for a prize. Just because the game presented a finite, numbered list and somebody had to find out.
They found out. Every one of them is winnable except deal 11982, which is provably impossible. That story - a game handing out a finite catalogue of puzzles and a community exhausting it - is unusual enough on its own, but what the solvers learned along the way is more interesting than the scoreboard.
What a FreeCell solver actually does
A solver treats the deal as a search problem. Each arrangement of cards is a state; each legal move is an edge to another state; the goal is any state with all four foundations complete. That is a graph, and solving a deal means finding a path through it.
The difficulty is that the graph is enormous. A mid-game FreeCell position routinely offers a dozen or more legal moves, and sequences run to a hundred moves or more. Exploring blindly is hopeless, so solvers lean on two things: a heuristic that ranks positions by how promising they look, and aggressive pruning of positions already seen.
The heuristics are recognisably the same advice given to human players. Reward empty columns heavily. Penalise occupied free cells. Reward cards sitting in long ordered runs. When the machine's scoring function and the human rule of thumb agree this closely, it is usually a sign the rule of thumb is genuinely correct rather than folklore.
Why the win rate is so absurdly high
The reason nearly every deal falls is structural. FreeCell gives you four columns of seven and four of six, all face up, plus four cells. That is a lot of working space relative to 52 cards, and crucially the space is flexible: any card can go to any cell, and an emptied column accepts anything.
Contrast a game where the working space is rigid. In Klondike the stock arrives in a fixed order you cannot influence, face-down cards gate whole columns, and a meaningful share of deals are dead from the start. If you play Solitaire online and lose, the honest answer is often that the deal was not winnable - which is a fundamentally different relationship with the game.
The other structural gift is that FreeCell's randomness is spent entirely at the deal. After that the game is deterministic and fully observed. Games with that shape - random once, then nothing hidden - are almost always more solvable than they feel, because the only obstacle is search depth rather than information. The tile version behaves the same way: play Mahjong online and the board is generated to be solvable, then everything is visible.
What 11982 actually looks like
The unwinnable deal is not obviously monstrous. It looks like a normal deal. What kills it is a configuration where the cards needed to free each other are mutually buried in a way that no ordering of four cells and eight columns can untangle - the aces and low cards are positioned such that every route to one costs access to another.
The instructive part is that no human eyeballing the layout would call it. It took exhaustive search to prove impossibility, because proving a deal unwinnable means showing that no path exists - you have to exhaust the graph, not just fail to find a route.
Which is a useful thing to know at the table: when you are stuck, "this is unwinnable" is almost never the right conclusion. The odds are roughly 32,000 to 1 against you, and the actual answer is nearly always that the dig you started was not planned deeply enough.
What this means for how you play
Three consequences follow directly from the solver results.
Undo is not cheating, it is search. A solver explores and backtracks. A player who never backtracks is doing depth-first search with no backtracking, which is the worst possible algorithm. If you are trying to actually beat a deal rather than test your first instinct, taking moves back is the correct method.
Your losses are sequencing, not luck. Because the deal was winnable, the failure was an ordering decision. The most common one by far is spending an empty column on a card that did not need to be there.
The hard deals are hard by depth, not by trick. The famously difficult numbered deals do not contain a special rule or a hidden mechanism. They simply require a longer forced sequence before any progress shows. That is why counting the whole dig before starting it matters so much.
This is the same skill that a game with bounded, countable randomness rewards. Spend an evening with Tetris online and the piece sequence is a shuffled bag rather than a free-for-all, so a strong player is quietly counting what must arrive - and the losses come from stack management, not from the draw. The parallel is exact: bounded randomness plus visible state equals a game where the excuse is not available.
The other end of the spectrum
It is worth occasionally playing something with the opposite contract, if only to appreciate what FreeCell offers. In a dice game the randomness never stops and correct play still loses - play Yahtzee online and the upper-section bonus will slip away on a run of bad rolls no matter how well you allocate. And in a trick-taking game the information is hidden but recoverable, which is a third category again: play Hearts online and the cards you cannot see shrink with every trick, so the skill is memory rather than search.
FreeCell sits alone: nothing hidden, nothing random after the deal, and a near-certainty that a solution exists. That is a rare and rather generous design.
Try it yourself
Take a deal you lost, and replay it with undo, treating it as a puzzle rather than a performance. Most players find the winning line within two or three attempts, which is the clearest possible demonstration of what the solvers proved at scale.
If you want the difficulty turned up honestly, Baker's Game uses the same board with same-suit building, and Seahaven Towers tightens the space further. The rules hub and the glossary cover the mechanics, and the winnability FAQ gives the short version of this whole piece.
For the strategy side - the move formula and why an empty column beats a free cell - see the companion article.
💡 Put it into practice: deal a free game of FreeCell, read the rules of every variant, or try today's daily challenge.