Causal Auditing of Harm Discrepancies in Patient World Models
October 6, 2026
Clinical trial simulation models can agree on risk parameters but disagree on the actual fraction of patients harmed. This research uses structural causal models (SCMs) and response-mismatch budgets to audit these reliability gaps in counterfactual reasoning.
HOW THIS AFFECTS YOU
●
researcherThe study defines exactness boundaries for sensitivity frontiers in two-stage shared-response SCMs.
●
healthYou must audit clinical AI for disagreements in counterfactual harm estimates, even if transition probabilities appear consistent.