Output Type Before Quality: A Standards-Derived XAI Admissibility Rubric for Autonomous-Driving Safety

The paper introduces a standards-derived rubric for evaluating explainable AI (XAI) methods in the context of autonomous driving safety, addressing the mismatch between safety standards and current XAI output types. It establishes 19 testable criteria across seven lifecycle stages, highlighting that causal XAI methods are essential for hazard identification, incident investigation, and data management, with a significant gap in structural admissibility for these stages. This framework emphasizes the need for practitioners to select XAI methods based on lifecycle-stage evidence requirements rather than popularity, thereby improving the reliability of safety assurances in autonomous driving systems.