Evaluating Bivariate Causal Statements Based on Mutual Compatibility

In three linesMethod to evaluate collections of bivariate causal statements without access to ground truth. Authors introduce a compatibility score quantifying plausibility of induced causal models, without relying on faithfulness assumption. Applied to assessing causal claims from large language models.

Summary generated by Claude — human-verified