Introduction

Cassville Checkers is a marble racing board game that has been played in my family for generations. The game is played on a handmade wooden board with a distinctive ring topology—marbles race around a circular track, competing to be the first to get all their pieces home.

This analysis site documents our exploration of optimal strategies for the game using reinforcement learning and heuristic agents.

The Game¶

Cassville Checkers is a 2-4 player game where each player has 5 marbles that must travel from a home base, around a circular ring, and into a goal area.

Board Layout¶

The board consists of:

• Home bases: Each player starts with 5 marbles in their home area

• Staging areas: A waiting zone before entering the main ring

• Circular ring: A 48-position track that all players share

• Goal areas: 5 positions where marbles finish their journey

Movement Rules¶

1. Deployment: Roll a 1 or 6 to move a marble from home to the staging area

2. Entering the ring: Any roll moves a marble from staging onto the ring

3. Ring movement: Move clockwise around the ring by the die value

4. Lap completion: Marbles must complete one full lap before they can enter the goal

5. Entering goal: Exact roll required to land on a goal position

Special Rules¶

• Capturing (“Zapping”): Landing on an opponent’s marble sends it back to their home

• Bonus turns: Rolling a 6 grants an additional turn

• Mercy rule: After 3 consecutive failed deployment attempts, you can bypass the 1/6 requirement

Winning¶

The first player to get all 5 marbles into their goal area wins.

Project Goals¶

This analysis aims to answer several questions:

1. What strategies work best? We compare random play, priority-based heuristics, and score-based greedy approaches.

2. Can reinforcement learning discover good strategies? We train PPO agents and evaluate their performance against heuristic baselines.

3. What insights emerge? Through extensive benchmarking, we identify key principles for effective play.

Analysis Overview¶

The following sections detail our findings:

• Implementation: Technical details of the Gymnasium environment and agents

• Strategy Analysis: Detailed benchmark results comparing different agent strategies

• Conclusions: Key insights and recommendations for optimal play