DungeonBench: Benchmarking Tactical Reasoning in D&D Combat
A new standard known as DungeonBench has been launched to assess tactical reasoning within Dungeons & Dragons combat scenarios. This benchmark encompasses nearly all combat-related material from the 2014 System Reference Document, preserving mechanics that are typically overlooked by simplified combat simulators. DungeonBench offers a comprehensive tactical overview at each stage, including a decision to be made and a detailed list of possible actions, such as movement, attacks, spells, reactions, objectives, preparations, and limited resources. The objective is to evaluate valid choices based on action economy, creature characteristics, battlefield layout, timing, and future conditions. This benchmark seeks to improve the testing of complex tactical reasoning in existing game and simulator evaluations. The paper can be accessed on arXiv under the identifier 2607.29577.
Key facts
- DungeonBench is a benchmark for tactical reasoning in Dungeons & Dragons combat.
- It covers the vast majority of combat-relevant 2014 System Reference Document content.
- The benchmark retains mechanics that simplified combat simulators often abstract away.
- At each step, it exposes a complete tactical observation, a pending decision, and an indexed list of executable options.
- Options span movement, attacks, spells, reactions, objectives, preparation, and scarce resources.
- The task is to value legal choices whose consequences depend on action economy, creature traits, battlefield geometry, timing windows, and future states.
- The benchmark addresses under-testing of rules-rich tactical reasoning in current game and simulator benchmarks.
- The paper is available on arXiv with identifier 2607.29577.
Entities
Institutions
- arXiv