Prometheus: Automating Deep Causal Research Integrating Text, Data, and Models

PROMETHEUS is a framework designed to enhance the utility of causal claims extracted by large language models by organizing them into navigable world models called causal atlases. These atlases consist of local causal predictive-state models that include structured claim tables, predictive tests, support statistics, and provenance. The framework allows for the comparison of overlapping regions and identifies areas of agreement, drift, contradiction, and underdetermination. PROMETHEUS is demonstrated through case studies in various fields, such as marine biology, weight-loss evidence, and health-benefit claims, as well as grounded-counterfactual studies using source data and simulations. This approach provides a research instrument for navigating and evaluating the strength and coherence of claims within a corpus.