FreeCell History: Origins, Early Rules and Evidence

Short answer: The strongest first-person evidence says that Paul Alfille programmed and probably named FreeCell on the University of Illinois PLATO system in the mid-1970s. He had already played a similar all-cards-visible solitaire with physical cards, but he could not remember the book or state exactly which rule he changed. Baker’s Game and Eight Off are important documented relatives, not a perfectly proven one-step genealogy. Jim Horne later encountered FreeCell on PLATO at the University of Alberta, wrote personal-computer versions, and placed his graphical edition into Windows builds. The famous result about deal 11982 concerns Microsoft’s particular 32,000 numbered-deal set, not every possible FreeCell deal SRC-025 SRC-026 SRC-033 SRC-036.

FreeCell feels as though it must have been designed for a computer. Every card is visible. Empty cells act like registers. Long transfers can be decomposed into smaller legal moves. A numbered shuffle can be replayed and analyzed by many people. Yet Paul Alfille’s own account begins not with software but with a remembered card game from childhood SRC-025. The computer did not create the underlying idea of an open packing game; it made that idea easier to deal, repeat, compare, automate and study.

That distinction is the key to a reliable history. There are at least four different questions: Which earlier games resemble FreeCell? What rules did Alfille remember or alter? When did he program the PLATO version? How did Jim Horne’s Windows edition turn one implementation into a familiar standard? Sources answer those questions with different levels of certainty. A publication date proves that a rule was in print by then. A creator interview records recollection, including uncertainty. A software release documents distribution, not the invention of the card-game family.

For the current PlaySoli procedure, read the complete FreeCell rules. For a direct structural comparison with the best-known hidden-card patience, see FreeCell versus Klondike.

Contents

• Why FreeCell history needs source discipline
• The older open-packing family
• Baker’s Game in 1968
• Eight Off and the limits of a neat family tree
• What Paul Alfille remembered
• FreeCell on PLATO
• What the computer added
• Jim Horne’s route to Windows
• The 32,000 numbered deals and game 11982
• What solver research does—and does not—prove
• Evidence-based timeline
• Common historical mistakes
• In brief
• Frequently asked questions

Why FreeCell history needs source discipline

Card-game histories are often presented as simple chains: game A became game B, one inventor changed one rule, and a later company made it popular. The surviving evidence is rarely that tidy. Games circulate orally. Different households use similar layouts under different names. A programmer may remember a childhood rule without remembering its printed source. Later historians then compare structures and propose a likely relationship.

For this guide, the evidence is ranked rather than flattened:

1. Paul Alfille’s 2000 interview is primary evidence for his memory, his PLATO work and the features he says he implemented SRC-025.
2. Jim Horne’s interview is primary evidence for how he encountered the game and wrote the Microsoft version SRC-026.
3. The University of Illinois history provides institutional context for PLATO SRC-027.
4. Martin Gardner’s June 1968 Scientific American record proves that his relevant “Mathematical Games” column appeared in that issue SRC-035.
5. Specialist and software documentation help compare Baker’s Game, Eight Off and FreeCell, but structural similarity alone does not prove Alfille’s exact childhood source SRC-033 SRC-036.
6. Academic solver papers document defined rule models and computational results; they do not convert a sample or solver score into a universal win probability SRC-037 SRC-038 SRC-039.

The phrase the earliest description we have located is therefore more accurate than “the first ever.” Likewise, “Alfille’s PLATO implementation” is safer than a claim that no one had ever played the same physical-card rules before him.

The older open-packing family

FreeCell belongs to a wider family sometimes called open packers: most or all cards begin face up, the tableau is reorganized under a packing rule, temporary spaces provide mobility, and foundations receive cards in suit order. Documented relatives such as Baker’s Game and Eight Off predate Alfille’s FreeCell program SRC-033 SRC-036.

The central design problem is easy to recognize. If every card is visible, uncertainty is reduced, but access is still constrained. A needed low card may sit beneath a high card. A temporary cell can free it, but occupying that cell reduces future transfer capacity. Clearing a column creates a much more powerful workspace, yet a careless card placed there can trap the space again. Those tensions appear in related games even when their exact building rules differ.

Calling a game an “ancestor” can mean several things. It may mean that the rules are structurally older. It may mean that a documented designer explicitly adapted them. Or it may mean only that later historians place the games in the same family. FreeCell’s history supports the first and third meanings more strongly than a perfectly documented line of direct transmission SRC-025 SRC-033 SRC-036.

Baker’s Game in 1968

Martin Gardner’s “Mathematical Games” column titled “Combinatorial possibilities in a pack of shuffled cards” appeared in the June 1968 issue of Scientific American SRC-035. Modern PySolFC documentation identifies the open game associated with that column as Baker’s Game and explains its key distinction from FreeCell: tableau piles build downward by suit, whereas standard FreeCell builds downward in alternating colors SRC-036.

Baker’s Game otherwise has the recognizable ingredients of the family: one deck, all cards dealt at the start, four one-card cells, foundations, and sequence movement limited by the temporary spaces needed to reproduce the transfer as single-card moves SRC-036. The same-suit packing rule creates tighter dependencies than alternating colors because each rank has only one legal parent rather than two opposite-color possibilities.

This is strong evidence that a FreeCell-like open packing structure was publicly documented before Alfille’s PLATO program. It is not, by itself, proof that Alfille had Gardner’s issue in front of him or consciously performed a single named modification. Alfille said he remembered reading about a similar solitaire as a child, but he could not identify the book and did not know which variation he might have introduced SRC-025. The responsible conclusion is therefore:

• Baker’s Game is a documented and closely related predecessor SRC-035 SRC-036.
• Alternating-color packing is a defining difference in standard FreeCell SRC-036.
• The exact path from Gardner’s column—or another book—to Alfille’s remembered physical-card game is not established by his interview SRC-025.

Eight Off and the limits of a neat family tree

Eight Off is another established member of the same open-packing family. As its name suggests, it uses more reserve cells than standard FreeCell. PySolFC describes it as a version related to King Only Baker’s Game with eight cells SRC-036. Specialist histories commonly place Eight Off, Baker’s Game and FreeCell in one lineage SRC-033.

The temptation is to draw a straight arrow from Eight Off to Baker’s Game to FreeCell. The rules do show a meaningful progression of related ideas: open information, suited foundations, reserve spaces and constrained sequence transfer SRC-033 SRC-036. But the documentary record used here does not establish that Alfille learned one named version, changed exactly one rule, and then published the result. His own recollection is explicitly less precise SRC-025.

A better diagram would show a family of related open packers, with Baker’s Game documented in 1968, Eight Off as an older related form, and Alfille’s physical-card recollection feeding into the PLATO implementation SRC-025 SRC-033 SRC-035 SRC-036. The arrows should be labeled “structural relationship” or “probable influence,” not “proven direct copy.”

What Paul Alfille remembered

In a phone interview conducted in 2000, Alfille said he was not sure exactly where the game came from. He remembered reading about solitaire games as a child and recalled one in which all cards could be seen, making the puzzle more like chess than a game dependent on hidden information SRC-025. He played a version with physical cards before programming it.

When asked about a supposed predecessor and the variation he introduced, Alfille did not claim certainty. He wanted to know which book the interviewer had in mind, said he thought he made up the name FreeCell, and acknowledged that he did not know what variant he may have added SRC-025. These are unusually valuable statements because they preserve the boundary between memory and proof.

They also complicate the word “invented.” Alfille can reasonably be credited with creating the PLATO FreeCell program and with establishing the named computerized form described in his interview. His recollection supports probable authorship of the name, but “I think” is not the same as a surviving dated naming document. And because he played a related physical-card game first, the program’s underlying mechanics did not appear from nothing SRC-025.

The exact year deserves similar care. The interview’s introduction places the PLATO version in the mid-1970s SRC-025. Many later summaries give 1978, but the primary interview reviewed here does not pin the implementation to that precise year. This package therefore uses “mid-1970s” when relying on the interview and treats 1978 as a later conventional dating rather than an independently established fact.

FreeCell on PLATO

PLATO began at the University of Illinois in 1960 as a computer-assisted learning system. By the early 1970s, improved plasma displays, networked access and increasingly capable authoring tools supported graphical educational programs, communication and games SRC-027. That environment mattered to FreeCell. It offered a shared computer, a graphical display and an audience able to replay and compare a deterministic puzzle.

Alfille was a medical student, not a card-game company employee or a computer-science professor. He had programming experience and wrote FreeCell in TUTOR, PLATO’s instructional authoring language SRC-025. He described constraints that now sound severe: global variables, no recursion as he remembered it, limited storage and a display whose available space shaped the range of layouts.

The program’s small footprint did not prevent ambitious features. Alfille said users could choose variants ranging from four to ten columns and from one to ten cells. The standard eight-column, four-cell form was one point in a configurable system rather than the only possible layout SRC-025. He also maintained winning streaks and competitive records. A central machine made shared rankings and selected challenges practical long before web-based leaderboards.

These details show why PLATO was more than a neutral porting platform. It turned a hand-dealt solitaire into a repeatable system with recorded performance, configurable rules and a community. The program could also automate obvious endgame moves and help determine when a position was lost, according to Alfille’s account SRC-025. Those are software affordances, not changes to the objective of moving cards to foundations.

What the computer added

The computer’s largest contribution was not simply animation. It changed what could be measured and repeated.

Reproducible deals

A shuffle generated from a number can be reconstructed. Players in different places can attempt the same deal, compare lines and distinguish “I did not solve it” from “no solution exists.” That distinction later became central to the Microsoft numbered set SRC-026.

Enforced move legality

A program can reject a sequence transfer when there are not enough temporary spaces to decompose it. It can also perform a legal multi-step transfer as one visible action. The screen move may look like a new rule, but its legitimacy rests on the underlying one-card moves; the planned supermoves guide will treat that mechanism in detail.

Records and competition

Alfille’s PLATO version tracked streaks and supported competitive play SRC-025. This did not turn FreeCell into a two-player card game. It created social comparison around independent attempts at the same type of puzzle.

Automated housekeeping

A program can send clearly safe cards to foundations, finish a forced endgame or undo a move. These conveniences vary between implementations. They should not be mistaken for universal historical rules.

Jim Horne’s route to Windows

Jim Horne’s account separates three stages. First, he encountered FreeCell while working with a PLATO system at the University of Alberta. He explicitly said he did not invent the game SRC-026. Second, he wrote a personal-computer version using character graphics and distributed it through CompuServe. Third, after joining Microsoft, he created a graphical Windows version.

Horne says he placed that version into a Windows NT build as a practical joke and later put FreeCell into the Windows 95 build SRC-026. Whatever the informal route into those builds, his account documents the transition from PLATO and character-based PC versions to a graphical edition included in Microsoft product builds.

This is why several roles must remain distinct:

• Paul Alfille created the PLATO implementation and probably the name, on his own qualified recollection SRC-025.
• Jim Horne created the Microsoft Windows implementation after learning the game through PLATO SRC-026.
• Microsoft product builds distributed Horne’s graphical implementation beyond PLATO and his earlier PC version SRC-026.
• Earlier open packers supplied related mechanics before either program SRC-033 SRC-035 SRC-036.

Calling Horne the inventor of FreeCell erases Alfille and the physical-card context. Calling Microsoft the inventor confuses distribution with design. Calling Alfille merely a porter, however, understates the named PLATO system, its configurable forms and its competitive features SRC-025 SRC-026.

The 32,000 numbered deals and game 11982

The Windows version’s numbered deals produced one of FreeCell’s best-known historical episodes. Horne says the original set contained 32,000 games generated reproducibly from their numbers. Players were assigned batches, reported unresolved deals, and progressively concentrated effort on the remaining cases SRC-026. Eventually one deal—number 11982—stood apart.

Later computational work established that 11982 is unsolvable under the defined Microsoft rules, while the other deals in that particular set are treated as solvable SRC-026 SRC-033 SRC-038. The result is important, but its scope must be stated precisely:

• It concerns the 32,000 shuffles produced by a particular generator and numbering scheme.
• It assumes a particular FreeCell ruleset and interpretation of legal moves.
• It does not prove that only one arrangement among all possible 52-card deals is unsolvable.
• It does not mean that an individual player should win 31,999 out of every 32,000 attempts.
• It does not justify a solvability claim for PlaySoli deals unless those deals and rules are independently matched and analyzed.

These scope limits follow from the defined Microsoft generator, Horne’s account and later benchmark descriptions SRC-026 SRC-033 SRC-038.

Horne also described deliberately constructed negative-number deals that were obviously impossible. That observation alone disproves the slogan “every FreeCell deal is solvable” SRC-026. PlaySoli therefore makes no universal guarantee about a deal without solver or curation evidence SRC-001 SRC-007.

What solver research does—and does not—prove

FreeCell became a useful research domain because it combines complete visible information with a large search space. Academic work has studied heuristic search, deadlock detection and optimal solutions SRC-037. Another paper reported a solver that completed 98% of the Microsoft 32K benchmark used in that experiment SRC-038. That number is a solver success rate, not a measured percentage of solvable deals.

The distinction is fundamental:

• A deal can be solvable even when a particular program fails to find a solution within its method or resource limit.
• A solver can complete nearly all cases in a selected benchmark without describing all possible shuffles.
• A human win rate includes errors, incomplete search and chosen strategy.
• A mathematical solvability rate asks whether at least one legal winning path exists under an exact ruleset.

Academic work also formalizes sequence transfers and deadlocks, helping explain why a move that looks locally useful can destroy the only route through a dependency cycle SRC-037 SRC-039. The research validates FreeCell as a serious planning problem. It does not provide a universal promise that the next board—or every board on any website—must be winnable.

Evidence-based timeline

Date or period	What the evidence supports	What it does not prove
Before 1968	Related open-card packing games, including Eight Off, were already part of the patience family SRC-033 SRC-036.	A direct, fully documented line to Alfille’s childhood rules.
June 1968	Gardner’s relevant Scientific American column was published; Baker’s Game is associated with it in specialist documentation SRC-035 SRC-036.	That Alfille used this exact issue as his source.
Mid-1970s	Alfille’s original FreeCell ran on PLATO at the University of Illinois SRC-025.	A precisely established day or, from this primary interview alone, the often-repeated exact year 1978.
PLATO era	The program used TUTOR, allowed configurable columns and cells, and supported streaks and competition SRC-025.	That every later implementation copied every PLATO feature.
Pre-Windows PC era	Horne wrote a character-based PC version after encountering the game on PLATO at Alberta SRC-026.	That Horne invented the underlying game.
Windows NT / Windows 95 era	Horne created a graphical Windows version and says he inserted it into those product builds SRC-026.	That Windows was the first computer implementation.
Numbered-deal investigation	The Microsoft 32K set was collaboratively tested; deal 11982 became the proven unsolvable exception within that set SRC-026 SRC-033.	A universal FreeCell solvability percentage.
2010s onward	Researchers published solver, search and deadlock studies using defined FreeCell models SRC-037 SRC-038.	That one solver’s benchmark score equals the game’s global winnability.

Common historical mistakes

Saying “FreeCell was invented in 1978” without qualification

The date is widely repeated, but Alfille’s primary interview used “mid-70s” and did not supply a precise year SRC-025. Write that the PLATO implementation dates to the mid-1970s; identify 1978 as a common later dating only when a source is supplied.

Claiming Alfille copied Baker’s Game and changed exactly one rule

The structural comparison is persuasive, but Alfille could not identify his childhood source or the exact change he introduced SRC-025. Treat Baker’s Game as a documented close predecessor, not a fully proven one-step transmission.

Calling Jim Horne the inventor of FreeCell

Horne himself said he encountered the game on PLATO and did not invent it SRC-026. His contribution was the PC and Windows implementation path.

Calling Microsoft the creator of FreeCell

Microsoft distributed Horne’s version. Alfille’s PLATO program and related physical-card games came earlier SRC-025 SRC-026.

Saying every deal is winnable

Alfille and Horne both discussed constructible or identified unsolvable arrangements SRC-025 SRC-026. The Microsoft 32K result is a bounded set, not a universal theorem.

Turning 98% solver coverage into 98% winnability

The academic figure describes one solver’s benchmark performance SRC-038. It does not say that the remaining 2% were all unsolvable, nor that 98% of arbitrary deals are solvable.

Treating visible cards as proof of an easy game

Complete information removes hidden-card uncertainty. It does not remove combinatorial difficulty, dependency cycles or irreversible loss of workspace.

In brief

• FreeCell has older relatives in the open-packing family, especially Baker’s Game and Eight Off.
• Gardner’s June 1968 column is a verifiable publication point for Baker’s Game SRC-035 SRC-036.
• Alfille remembered a similar physical-card game but could not identify the exact source or variation SRC-025.
• He programmed FreeCell on PLATO in the mid-1970s, using TUTOR, and probably coined the name according to his own qualified recollection SRC-025.
• The PLATO version included configurable layouts, streak records and competition SRC-025.
• Horne encountered FreeCell on PLATO at Alberta, wrote PC versions and then a graphical Windows edition SRC-026.
• Deal 11982 is the unsolvable exception in Microsoft’s specific 32,000-deal set, not proof about every possible deal SRC-026 SRC-033.
• Solver performance, human win rate and mathematical solvability are different measurements SRC-037 SRC-038.

Frequently asked questions

Who invented FreeCell?

Paul Alfille is best credited with creating the named PLATO FreeCell implementation in the mid-1970s. He had already played a related physical-card game and was unsure of its source, so the deeper rule ancestry is not attributable to one person with equal certainty SRC-025.

Was FreeCell invented in 1978?

1978 is a common modern date. The primary Alfille interview used the broader phrase “mid-70s,” so this package does not present the exact year as conclusively established by that interview SRC-025.

Did FreeCell come from Baker’s Game?

Baker’s Game is a closely related documented predecessor, distinguished chiefly by same-suit tableau building in the standard description SRC-036. The likely relationship is strong, but Alfille could not confirm which book or named game he remembered.

How is Eight Off related to FreeCell?

Eight Off belongs to the same open-packing family and uses eight reserve cells. Its shared ideas help explain FreeCell’s design context, but “ancestor” should not be mistaken for a fully documented chain of direct copying SRC-033 SRC-036.

Did Paul Alfille name the game?

He said, “I think I made up the name,” which is useful first-person evidence but still a qualified recollection rather than a dated naming record SRC-025.

What did PLATO add to FreeCell?

PLATO supplied a graphical, networked environment in which Alfille could automate dealing, offer configurable column/cell counts, keep streak records and support competitive challenges SRC-025 SRC-027.

Did Jim Horne invent Microsoft FreeCell?

Horne created the Microsoft Windows implementation. He had encountered FreeCell on PLATO and explicitly said he did not invent the underlying game SRC-026.

Why is deal 11982 famous?

It is the proven unsolvable deal in Microsoft’s original numbered set of 32,000 reproducible shuffles under that ruleset SRC-026 SRC-033. It does not imply that all other possible FreeCell arrangements are solvable.

Are nearly all FreeCell deals solvable?

Evidence from famous numbered sets and solver work suggests that many standard deals are solvable, but no percentage should be applied universally without defining the generator, rules and proof method. PlaySoli does not guarantee every deal SRC-007.

Related PlaySoli guides

• FreeCell rules: layout, legal moves and foundations
• FreeCell strategy: planning with cells and empty columns
• Common FreeCell mistakes and how to diagnose them
• FreeCell supermoves and transfer capacity
• FreeCell versus Klondike
• How computer solitaire developed

Sources used

• SRC-025 Denny Cronin, interview with Paul Alfille.
• SRC-026 Interview with Jim Horne on the origin of Microsoft FreeCell.
• SRC-027 University of Illinois Grainger College history of PLATO.
• SRC-033 Michael Keller, FreeCell FAQ and Links.
• SRC-035 Martin Gardner, “Mathematical Games,” Scientific American, June 1968, publication record and DOI.
• SRC-036 PySolFC documentation for Baker’s Game and Eight Off.
• SRC-037 Gerald Paul and Malte Helmert, research on optimal FreeCell solutions and deadlock analysis.
• SRC-038 Achiya Elyasaf, Ami Hauptman and Moshe Sipper, GA-FreeCell.
• SRC-039 Marten Klaver, Theoretical and Practical Aspects of Freecell.

Material checked: 2026-07-17
Disputed or unverified facts: The exact printed game Alfille remembered, the precise rule change he introduced, a definitive year beyond the primary interview’s “mid-1970s,” and a universal solvability percentage remain unproven or scope-dependent.
Editorial responsibility: PlaySoli Editorial Team.