Solitaire Poker, released for the Sega Game Gear, is a card-based puzzle game in which the player builds poker hands across a grid to maximize the score. In a normal playthrough, the goal is to score enough points by forming hands to advance through the game. In this TAS, I've made it that the goal is to complete the run as fast as possible while still reaching a strong final score.
This project is an interesting optimization challenge because it combines puzzle solving with frame-level timing. To tackle it, I developed a custom Python-based solver that models the board as a search space of possible moves and evaluates partial states using poker-hand scoring heuristics. The solver uses a branch-and-bound strategy to explore promising move sequences while pruning weaker branches, allowing it to focus on the most efficient solutions.
Game objectives
- Finish all 15 levels of the default one-player View-1 mode in the minimum number of frames
- Afterward, maximize the final score of the completed boards
Approach
I built a custom search tool tailored to this game’s structure. The program evaluates rows, columns, and diagonals as poker hands, then uses a branch-and-bound search to explore the most promising move sequences while discarding branches that cannot beat the current best upper bound. I also added heuristics and caching to make the search practical and guide it toward strong configurations. The implementation uses the treys Python package for fast poker-hand evaluation.
The game uses two areas: a source area on the left, where cards are selected from columns, and a play area on the right, where cards are placed to form poker hands. To optimize for time, the main constraint is which column can be chosen from the source area.
Possible improvements
- Further frame savings may be possible by better understanding the level-to-level frame "randomness" that is still outside my current understanding.
- I don't think in-level timing can be improved much further.
- A higher-score solution is possible with additional search time or a better search configuration.
This was a fun and rewarding project to work on, and I am always happy to apply algorithmic search methods to video games. Unfortunately, the final result is so fast that it is difficult to follow visually. Still, I hope this TAS can serve as a solid starting point for future optimization work on this title.
Code repo
