The Win-Rate Observatory
Solitaire Win Rates: Winnable vs Actually Won
What percentage of solitaire games are winnable? For classic Klondike, about 79–82% of deals can be won in principle — but across roughly 188,000 real Klondike games in our June 2026 historical baseline, players actually won 31%. That gap — between what a perfect player could do and what real hands actually do — is what this page measures, for every solitaire variant we run: win rates from our own telemetry beside solver-computed ceilings, with the full methodology below and the data free to reuse with attribution.
Every game we run, measured and bounded
| Game | Won by players here | Winnable in principle | Notes |
|---|---|---|---|
| Klondike (Draw 1) | 31% 188,000 games, June 2026 | ≈ 79–82% literature | unlimited undo included |
| Spider — 1 suit | 50.0% 2,257 games, June 2026 | Practically every deal literature | |
| Spider — 2 suits (June Random) | 10.4% 1,195 games, June 2026 | Published verified pool: 90 of 90 exact 90 deals | Deals here are pre-verified winnable (curated pool) |
| Spider — 4 suits | Logging since July 2026 | Most deals — if the line can be found literature | |
| FreeCell | 11.1% 1,234 games, June 2026 | 31,999 of 32,000 (99.997%) catalog 32,000 deals | the original Microsoft catalog; the modern 1,000,000-deal catalog is ≈99.999% |
| TriPeaks | Logging since July 2026 | 96.6% exact 1,000 deals | |
| Pyramid | Logging since July 2026 | 65.8% exact 1,520 deals | Deals here are pre-verified winnable (curated pool) |
| Golf (strict rules) | Logging since July 2026 | ≈ 27% exact 2,000 deals | Deals here are pre-verified winnable (curated pool) |
| Yukon | Logging since July 2026 | ≈ 75% lower bound | solver finds a full line on roughly three of four random deals within budget |
| Alaska | Logging since July 2026 | ≈ 80% lower bound | solver finds a full line on about four of five random deals within budget |
| Scorpion | Logging since July 2026 | ≥ 12.8% lower bound 1,000 deals | 224 of 1,000 deals undecided at the search budget |
| Wasp | Logging since July 2026 | ≥ 77.9% lower bound 1,000 deals | |
| Forty Thieves | Logging since July 2026 | ≥ 7.7% lower bound 1,000 deals | 923 of 1,000 deals unresolved by the bounded search (922 hit the 2M-node budget; 1 ended without a winning line); avg best line banks 40.6 of 104 |
| Crescent | Logging since July 2026 | ≥ 45% lower bound 1,000 deals | with all 3 shuffles; per-shuffle floors 5.8/13.0/26.9/45.0% and unresolved counts 942/870/731/550 at the bounded search limits; avg best line banks 102.2 of 104 |
| Canfield | Logging since July 2026 | ≥ 33.1% lower bound 1,000 deals | 441 of 1,000 proven lost within the 3-redeal search cap (play allows unlimited); 228 undecided at the 2M-node budget; avg best line banks 27.4 of 52 — $136.75 at the 1890s casino payout |
| Eight Off | Logging since July 2026 | ≥ 97.7% lower bound 1,000 deals | 0 of 1,000 proven lost; 23 undecided at the 2M-node budget; avg best line banks 51.5 of 52 |
| Russian | Logging since July 2026 | ≈ 7% exact 2,000 deals |
Measured column: methodology v1 historical-baseline, unlimited undo included, snapshot dated 2026-06-30. It is not blended with the active corrected v2 series, which is collecting a mature cohort. Full definitions in the methodology below; the same numbers ship as open JSON.
Winnable versus won: why the gap is so large
"Winnable" means a winning line exists for a player who somehow knows every card — including the face-down ones — and never wastes a move. Real solitaire is played against hidden information: you commit to plays before the cards they depend on are revealed, and one early bank sent up too soon can quietly strand the game ten moves later. That is why the two numbers on this page tell different stories, and why the gap is widest exactly where recovery depends on foresight: nearly every FreeCell deal is provably winnable, yet our players win about one game in nine. Neither number is wrong. The ceiling describes the deal; the measured rate describes the human experience of playing it.
How we measure real win rates
The measured figures currently published here are a frozen methodology v1 historical baseline: GA4 event-count snapshots from June 2026, calculated as games won divided by deals started. They are useful history, but the old instrumentation could miss initial attempts and count a win again after undo, so we do not present them as corrected attempt-level estimates. Methodology v2 started at the explicit production cutover on 15 July 2026 at 12:50:26 UTC: BigQuery joins each mature start to zero or one observed terminal outcome through a stable pseudonymous attempt ID namespaced by GA’s pseudonymous user ID, deduplicates event IDs within that namespace, and requires methodology_version=2. A start whose browser disappears without a terminal remains in the denominator as not won. V1 and v2 are separate series and are never blended. Practice boards and demos stay excluded; only analytics-granted sessions are measured, whether through an explicit choice or a verified location outside our configured GDPR-region set.
How the solvability ceilings are computed
The ceilings come in four honest flavors, and the table badges each one. Exact: our solver searched every reachable line of every sampled deal to a verdict — TriPeaks 96.6% of 1,000 random deals, strict Golf about 27% of 2,000, Russian solitaire about 7% of 2,000, Pyramid 65.8% of 1,520. Lower bound: the solver proved wins within a search budget and ran out of budget on the rest, so the true share is higher — Scorpion at least 12.8%, Wasp at least 77.9%, Yukon roughly three of four, Alaska about four of five. Catalog: FreeCell’s original 32,000 numbered deals have been settled for decades — 31,999 are winnable. Literature: for Klondike we quote the published research estimate of 79–82% for thoughtful Draw 1 play, because we do not yet run a Klondike solver of our own — and the badge says so. Where a repo benchmark script backs a figure, the methodology links it.
Reading the table honestly
Three caveats keep these numbers truthful. Self-selection: our Draw 3 players win about as often (33%) as Draw 1 players (31%) — not because Draw 3 is easier, but because experienced players choose it. Curated pools: on our Golf and Pyramid pages every deal is pre-verified winnable before it reaches you, so measured rates there describe play on winnable deals only and cannot be compared to a random-deal ceiling — those rows are labeled. Young pages: games we launched recently show "logging since July 2026" instead of a rate; we publish a measured figure once roughly a thousand real deals are logged, and never below five hundred.
Solitaire win-rate FAQ
What percentage of solitaire games are winnable?
For classic Klondike, research puts thoughtful Draw 1 play at about 79–82% winnable. Other variants span a huge range: nearly every FreeCell deal is winnable, TriPeaks reaches 96.6%, while Russian solitaire bottoms out around 7%. The full table above lists every variant we run — and the more surprising number beside it: what players actually win.
What percentage of solitaire games do players actually win?
On this site, June 2026: Klondike 31% across roughly 188,000 games, Spider 50.0% / 10.4% / 1.1% by suit count, FreeCell 11.1%. The 10.4% figure is the historical two-suit Random result; verified deal tiers are measured separately. These are real human results with unlimited undo included — measured from our own telemetry, with sample sizes published, not estimates.
What is the hardest solitaire game?
By solvability ceiling, Russian solitaire: our solver settled 2,000 random deals exhaustively and only about 7% can be won by any line of play. By measured human results, 4-suit Spider is the brutal one — 1.1% won here. Scorpion sits between: at least 12.8% solvable, with the true share unknown because some searches exhaust their budget.
Is every FreeCell game winnable?
Essentially — 31,999 of the original 32,000 numbered deals are provably winnable. Our players still win only 11.1% of FreeCell games, which is the cleanest demonstration on this page that a lost game is almost always a recoverable mistake rather than a bad deal.
Why are your numbers lower than other sites publish?
Most published solitaire "win rates" are theoretical solvability figures or unsourced folklore. Ours are dated telemetry snapshots with sample sizes shown; the currently published figures are labeled methodology v1 historical baseline, and corrected methodology v2 will remain a separate attempt-level series. Real play against hidden information loses to the theoretical ceiling everywhere — that gap is the finding, not an error. The dataset is free to reuse with attribution.
How often are these numbers updated?
Deliberately, not silently — each figure carries the window it was measured in, and we refresh the snapshot every few months or when a newly launched game crosses the publishing threshold. The open JSON endpoint linked below always carries the same numbers as this page.
Cite or embed this data
The charts, the table above, and the open JSON dataset are licensed CC BY 4.0 — free to republish, remix, and hotlink, with one condition: credit this page with a link (details in our copyright note). Everything below copies ready to paste:
<a href="https://playsolitaire.io/solitaire-win-rates">
<img src="https://playsolitaire.io/charts/solitaire-win-rates.svg"
alt="Solitaire win rates: winnable in principle vs actually won (playsolitaire.io, CC BY 4.0)"
width="860" loading="lazy" />
</a>
Prefer the difficulty ladder? Swap in /charts/solitaire-difficulty-ladder.svg.
Text citation: PlaySolitaire.io Win-Rate Observatory, https://playsolitaire.io/solitaire-win-rates (CC BY 4.0), data as of 2026-06-30, methodology v1 historical baseline.