Search over DAGs guarantees d-separation factorization, not the correct causal graph

fdellaert · x · 2026-09-21

Replying to a discussion about a causal inference notebook: the distribution is factorized over the selected graph's d-separation by construction, since the notebook searches over candidate DAGs—but that doesn't mean it identifies the right causal graph.

Related event: LLMs can draft DAGs but can't guarantee d-separation(2 posts)→

Original post →

More from Research

Research channel →