Yang Ni's review, posted on arXiv (2610.23543), argues that causal discovery has historically leaned primarily on conditional independence, but modern methods exploit properties of distributions beyond that. The review's organizing principle separates the source of identification—what makes causal structure learnable—from the statistical estimator and its computational implementation. That framing connects conditional-independence methods with distributional asymmetries, conditional-moment restrictions, and variation across environments or individuals.

The review also examines discovery under latent confounding and feedback. A key distinction is drawn between causal information about observed variables and reconstruction of latent structure; cyclic models, the review notes, require additional semantics beyond directed acyclic graphs. Ni separates population identification, frequentist and Bayesian consistency, uncertainty quantification, and algorithmic guarantees, and gives examples where the same observational distribution supports different conclusions under different model classes.

Because there is only one source, there are no disagreements to compare. The review concludes with guidance for comparing methods at their identifiable targets and a research agenda covering uncertainty, heterogeneous populations, non-scalar variables, and the intersection of discrete data, feedback, and hidden confounding.