HumanStudy-Hub/HumanStudy-Bench

Combines an Execution Engine that reconstructs full experimental protocols from published studies with standardized metrics that evaluate agent alignment at the level of scientific inference, not just task completion. Decouples base model capabilities from agent design choices, enabling precise attribution of results in human-subject simulation tasks. Community-driven with structured contribution workflows—studies are defined via JSON schemas and Python scripts, verified locally, and submitted as pull requests targeting a growing reference library.