RISE: Pragmatic Evaluation of Human-AI Reliance Calibration

Existing AI evaluation remains dominated by performance-centric benchmarks, limiting assessment of whether humans are using AI in calibrated ways. This paper proposes the Reliance in Intelligent Systems Evaluation (RISE) framework as a pragmatic reporting and interpretation standard for human-AI reliance calibration. A structured scoping synthesis of 27 relevant studies consolidated reliance indicators into an input-mediator-output structure comprising five families: appropriateness, raw reliance, verification/monitoring, emergent states, and inputs/outputs. RISE centers on a minimum reporting standard combining a core calibration panel with an explanatory overlay of trust, confidence, workload, and verification/monitoring. The result is a proposed reporting template for comparing designs and supporting human-centered AI test and evaluation.