Solitaire simulator hunts for best Klondike strategy — current record 8.590%

What’s new

A small, open-source simulator on GitHub is chipping away at one of solitaire’s quieter challenges: what's the best deterministic play order for Klondike? The project (https://github.com/dacracot/Klondike3-Simulator) runs millions of hands and applies simple move heuristics to see which sequencing wins more often. It has been reported that the current record win rate from this minimalist strategy set is 8.590%.

The headline is slightly absurd and oddly thrilling. You don’t need neural networks or Monte Carlo tree search here — just order the moves differently. In version 1.0 the simulator used the move sequence {s2g, b2g, b2b, s2b} and averaged about 7.915% wins. Changing the sequence to {s2g, b2b, b2g, s2b} in version 1.2 bumped the rate to 8.590%. That tiny tweak — a shift in priority — delivered roughly an 8.5% relative improvement. Sometimes the game is mostly about the order you try things, not deep foresight.

How to try it

Building and running is old-school Java tooling: run ant to compile, then java -jar simulator.jar with flags. Use --one or --three to set the draw, --attempts for number of games, --seed for repeatable shuffles, and --debug for verbose output. For example: java -jar simulator.jar --three --attempts 1000000 --seed 1111 will run a million hands; it has been reported that one million games takes less than an hour on an M2 MacBook Air. Want to beat the 8.590% mark? Fire up the simulator and see if a different tie-break rule gives you bragging rights.

This is a tidy little reminder: in some problems, brute-force simulation plus clever bookkeeping beats fancy models. Who knew solitaire would become a testbed for surgical, low-cost optimization?

Sources: github.com/dacracot, Hacker News