What is causal discovery evaluation and how is it validated?

Causal discovery evaluation measures how accurately a model infers causal graphs from data.

Causal discovery aims to reconstruct causal structures (DAGs) from observational or interventional data. Evaluation is challenging because ground truth causal graphs are rarely known in real-world settings. Common validation methods include synthetic benchmarks, intervention consistency checks, structural Hamming distance (SHD), and invariance tests across environments. A key challenge is distinguishing correlation from true causation, especially in high-dimensional systems.