EviDAG: Auditable Causal DAG Authoring with Biomedical Literature

EviDAG is a browser-based system for constructing biomedical causal DAGs as auditable, evidence-linked artifacts. It creates reproducible literature snapshots from free-text study concepts, uses an LLM module for structured pairwise causal judgments, attaches supported edges to verbatim evidence excerpts, and assembles a constraint-checked graph. The interface includes evidence review, graph comparison, adjustment-set computation, and export. The abstract reports high edge recall on literature-derived reference DAGs while preserving evidence trails missing from LLM-only baselines, but provides no numerical results or implementation details.