Tiny Moves: Game-based Hypothesis Refinement

TL;DR

This paper introduces The Hypothesis Game, a symbolic framework enabling LLM agents to iteratively refine scientific hypotheses through localized, domain-grounded moves, improving error correction and interpretability.

Contribution

It presents a novel game-based formalism for hypothesis refinement that emphasizes incremental reasoning, demonstrating its effectiveness over traditional prompting methods in scientific discovery tasks.

Findings

Game-based approach reduces more errors than baselines in corruption recovery.

Achieves higher precision while preserving valid hypothesis structure.

Performs comparably to top baselines in partial cue reconstruction tasks.

Abstract

Most machine learning approaches to scientific discovery frame hypotheses as end-to-end predictions, obscuring the incremental structure of scientific reasoning. We propose The Hypothesis Game, a symbolic formalism for hypothesis refinement in which LLM agents operate on a shared hypothesis state using a fixed grammar of reasoning moves. The framework is motivated by the observation that scientific progress often proceeds through small, localized revisions, grounded in domain context, rather than extensive rewrites. We instantiate a minimal game with LLM agents and evaluate it on pathway-level mechanistic refinement tasks. In the primary setting of corruption recovery, where hypotheses contain controlled errors, the game-based approach consistently removes more errors and achieves higher precision than strong prompting baselines, while preserving valid structure through incremental edits. In a secondary reconstruction setting from partial cues, it performs comparably to the strongest baseline, indicating that explicit move-based refinement remains competitive even when ground-truth recovery is difficult. These findings support game-based reasoning as a principled route to more controllable, interpretable, and transferable hypothesis refinement systems for scientific discovery.