Reasonable Doubt and Appellate Review Through Probabilistic Causal DAGs

Natural-language verdicts cannot reliably satisfy the reasonable doubt standard when many causal variables are at issue. The graph-theoretic conditions for ruling out alternative explanations are non-obvious to unaided verbal reasoning. We propose formalizing legal proof through probabilistic causal directed acyclic graphs (DAGs) grounded in Pearl's structural causal model framework, addressing the Dawid-Faigman-Fienberg criticism through a focus on appellate review of judgment-weight consistency. Reasonable doubt is defined as an active alternative causal path the evidence has not d-separated; the beyond-reasonable-doubt threshold is anchored in the Probability of Necessity. Criminal proceedings impose a path-closure obligation on the prosecution, while civil proceedings require a path-weighting competition between competing DAGs - structurally distinct tasks carrying different obligations. We further formalize Wright's NESS test within structural causal models and provide an edge-weight elicitation protocol grounding every weight in traceable evidentiary passages.