Alaska Solitaire: the widest gap between quick play and real planning
Alaska takes Yukon's layout — no stock, every card on the table, any face-up card movable with everything on it — and changes how you build: same suit only, one rank up or down. Every link you make is a same-suit link of the kind Spider players chase. The catch is that each card has very few places to go, and that is exactly where planning pays.
Of the games we tested for this site, Alaska shows the biggest difference between playing a little ahead and playing the whole game out. Our bot, which searches 300 positions ahead on each turn using only visible cards and never undoes, won just 1.5% of 10,000 deals (95% range 1.3% to 1.8%). A deep planning search, allowed to know the face-down cards, proved 75.6% of 500 deals winnable (95% range 71.6% to 79.2%). More than 74 points separate them. We did not run a plain greedy bot on Alaska, so short look-ahead is the lowest rung we can show.
Alaska at a glance
| What we measured | Result | n | 95% range |
|---|---|---|---|
| Short look-ahead bot (visible cards, 300 positions per turn, no Undo): won | 1.5% | 10,000 deals | 1.3–1.8% |
| Deep planning search (knows hidden cards): proven winnable | 75.6% | 500 deals | 71.6–79.2% |
| Deep search: undecided within budget | 17.2% | 500 deals | 14.1–20.8% |
| Deep search: proven impossible | 7.2% | 500 deals | 5.2–9.8% |
| Cards the bot got home, all games | 5.1 of 52 | 10,000 deals | 4.9–5.2 |
| Moves in the bot's wins | 110.9 | 151 wins | 108.9–112.9 |
Between 75.6% and 92.8% of deals can be won with perfect knowledge. The bot's 1.5% is a floor that shows how unforgiving Alaska is without planning or Undo; it is not how often people win. Blake and Gent's study does not cover Alaska, so we have no published figure to set beside ours.
Alaska rules on spidersolitaireonline.us
- Layout: seven columns: one card in the first, six to eleven in the others, the last five of each face up.
- No stock: all 52 cards are dealt to the tableau.
- Foundations: four, by suit, Ace to King.
- Building: same suit, one rank higher or lower. The 7 of hearts can take the 6 or the 8 of hearts.
- Group moves: pick up any face-up card and everything on top comes too. Only the card you lift must fit.
- Empty columns: a King only, with whatever sits on it.
What thinking ahead buys in Alaska
Same-suit links are worth protecting
A card has at most two tableau spots in the whole game: the cards one above and one below it in its own suit. When you build a same-suit link, you are using one of those scarce spots. Plan which link each card really needs before you spend it.
Use both directions
Up-or-down building doubles your options for any single card. If the 5 of spades cannot go on the 6, check the 4. Players moving on autopilot often forget the upward option.
Hidden cards are the real enemy
The bot ended with about 5 cards home on average across all games, usually stalled with the cards it needed still face down. Moves that turn a card over should come before moves that only tidy the face-up cards.
Don't read too much into the Aces
On average 1.6 of the 4 Aces start face down. Oddly, the bot won slightly more often when two or more were hidden (1.9%) than when none or one was (1.1%). We report it, but would not build a strategy on it; the order of your moves matters much more.
Plan for a long game
The bot's rare wins took about 111 moves. Alaska rewards patient, multi-step plans in which cards move more than once.
Let Undo do the deep search
The gap from 1.5% to at least 75.6% is what full planning is worth here. You cannot see face-down cards, but unlimited Undo lets you try a group move, see what it uncovers and take it back.
Features here
Alaska is free in any modern browser on computer, tablet or phone, with no download, account or email. Large cards and a zoom control, Hint, unlimited Undo, Restart, Auto Play, drag-and-drop or tap-to-move, and a tracker for wins, time and moves are included. Sound, fullscreen and table color are in the settings.
How we ran the study
The spidersolitaireonline.us team simulated Alaska in September 2026, alongside Yukon, on the rules this site's own game code uses.
- Read the game code: Yukon's deal and group moves with building changed to same suit, one rank up or down; Kings only in empty columns.
- Wrote the game in Python, sharing one program with Yukon with the building rule switched.
- Dealt numbered hands from seeds: 1 to 10,000 for the bot, 1 to 500 for the deep search.
- Used two players. The bot looks up to 300 positions ahead each turn using visible cards only, scoring cards home, hidden cards and good or bad links, and never undoes. The deep search is depth-first, knows every card, and had up to 50,000 positions per deal with no automatic foundation moves.
- Checked for published figures. Blake and Gent (2026) do not cover Alaska.
- Gave the uncertainty as 95% ranges: Wilson score intervals for percentages, normal intervals for averages.
- Stated the limits. The bot is weak at this game, the search sees hidden cards, and 17.2% of search deals stayed undecided.
The code uses only the Python standard library and ran across 19 processes: just over a minute for the bot's 10,000 games and about three and a half minutes for the search. Claude, an AI assistant made by Anthropic, helped write and review the code and draft this page. Every figure comes from running the code, not from the AI.
Alaska questions
Alaska or Yukon: which rewards planning more?
Alaska. Our bot won 1.5% here and 4.5% at Yukon, yet within the same search budget more Alaska deals were proven winnable (75.6% against 67.0%).
Why is Alaska hard when most deals can be won?
Each card has very few places to go and many start hidden, so winning takes a plan several moves long.
Are some deals impossible?
Yes. Our search proved 7.2% impossible even seeing every card.
Is it good practice for Spider?
Yes. Both reward keeping cards in suit. Try four-suit Spider for the toughest test, or browse the games page.
By Admin, spidersolitaireonline.us (Solitairen OOD). Research run September 2026.