This includes scene reconstruction, implementation of safety-defined attentive-human benchmarks, human and other road-user response models, counterfactual simulation following intervention, controlled scene variation, outcome and severity estimation, and the validation needed to understand when a result is credible enough to support a decision. Set the technical strategy and architecture for counterfactual behavior evaluation, including operationalizing safety-defined attentive-human benchmarks, counterfactual simulation following human intervention, alternate-policy comparisons, and controlled scene perturbations.