Technical Report for Splendor

This report describes a heuristic solver for Splendor that uses a gold-weighted evaluation function to guide decision-making.

Approach

The heuristic assigns higher value to gold tokens due to their flexibility as wildcards. The evaluation function considers:

1. Card accessibility — how many cards the player can currently afford
2. Noble proximity — distance to attracting noble patrons
3. Engine value — the discount value of owned development cards

def evaluate(state, player):
    card_score = sum(affordable_cards(state, player))
    noble_score = noble_proximity(state, player) * 2.0
    engine_score = discount_value(state, player) * 1.5
    return card_score + noble_score + engine_score

Results

The gold-weighted heuristic provides a strong baseline that significantly outperforms naive strategies while remaining computationally inexpensive.

Conclusion

Gold prioritization is an effective heuristic anchor for Splendor. The approach serves as a baseline for subsequent RL and LLM experiments.