Counterfactual Agentic Perturbation-Evidence Validation for V2X Safety Alerts

This work proposes CAPE-V2X (Counterfactual Agentic Perturbation-Evidence Validation for V2X), a semantic validation framework that verifies whether a received safety claim is physically supported by surrounding traffic behavior. Naturalistic high-D trajectory data are used to establish a behavioral prior for normal and disturbed traffic conditions, while controlled SUMO–Veins simulations generate legitimate and malicious V2X accident-alert scenarios. For each received claim, CAPE-V2X evaluates traffic behavior before and after the event, predicts the expected no-event traffic evolution, and measures the unexplained counterfactual residual. The system incorporates an adaptive threshold mechanism to improve performance under varying traffic conditions.