ExAtlas framework links social experiments, resolving conflicts with 98.6% accuracy

Researchers from the University of Chicago and collaborators have developed ExAtlas, a framework that turns the vast, disconnected archive of social and behavioral science experiments into a coherent atlas. By treating each study as a data point on a treatment-outcome surface, ExAtlas finds prior experiments locally close to a target study and checks if their effects can be composed to predict the target's result. If composition matches, the target is linked to consistent evidence. If it disagrees, ExAtlas reconciles the conflict by suggesting candidate moderators or higher-level theories. If composition fails entirely, it proposes new bridge experiments to fill the gap.

In tests on held-out targets with local support, ExAtlas correctly recovered effect direction in 98.6% of cases. Human evaluations further confirmed that its proposed bridge experiments are plausible and well-connected, and its conflict explanations help generate new theoretical insights. The work suggests that the existing experimental literature contains far more latent structure than is currently extracted, and making this structure explicit could guide both future theory and experimentation. The framework has applications in meta-science, evidence-based policy, and cumulative knowledge building in the social sciences.