RevengeBench: Reverse Engineering Code-Space Policies from Behavioral Experiments
2026-06-24 • Machine Learning
Machine Learning
AI summaryⓘ
The authors explore whether it's possible to figure out a player's hidden decision-making code just by watching their behavior in games. They created RevengeBench, a set of game challenges where AI models watch players and design smart tests to learn their strategies, then try to recreate the players' decision programs as runnable code. Their results show that AI models can partly recover these hidden strategies, especially when they can actively test opponents, helping weaker AIs improve. This work suggests it is possible to understand and predict behavior by reverse-engineering decision rules from observations and experiments.
inverse problempolicy reconstructionbehavioral probesLLM (large language model)executable codegame environmentElo ratingopponent modelinginterpretabilityaction-distance metrics
Authors
Babak Rahmani, Sebastian Dziadzio, Joschka Strüber, Sergio Hernández-Gutiérrez, Matthias Bethge
Abstract
For most of scientific history, researchers studying behavior could only infer hidden mechanisms from outward actions: an inverse problem that becomes more tractable when observation is augmented by targeted intervention. We pose a computational analogue: given only behavioral traces of an agent in a game environment, can a learner reconstruct the underlying decision program as executable code, and how much does this reconstruction improve with the ability to design controlled experiments? We introduce RevengeBench, a benchmark of 75 LLM generated, Elo-calibrated policies across five game environments, drawn from CodeClash tournament trajectories. The learner observes the hidden target policy play against sampled opponents and designs behavioral probes in the form of custom opponent policies that elicit informative behavior. It then submits an executable hypothesis, which is evaluated using continuous action-distance metrics. We further validate that recovered code carries informative signal in downstream player-versus-player tournaments. Across twelve frontier LLMs, recovery quality varies substantially (34 to 72% of initial distance closed), with reconstructed policies yielding measurable competitive advantage, particularly for weaker models that otherwise struggle to design effective counter-strategies. Our benchmark positions behavioral recovery of programmatic policies as a tractable inverse problem in code-space, opening a path to opponent modeling, policy interpretability, and the broader question of inferring latent mechanisms from observations.