This working paper introduces the Mentim framework, a structural approach to evaluating cognitive coherence and accountability in advanced AI systems. The paper formalizes the Unified Annotation Schema (UAS) and proposes the Structural Accountability Index (SAI), a composite metric designed to measure representational stability across multiple cognitive dimensions. The framework defines measurable primitives across Affective Energy, Ethics, Temporal Coherence, Transition Volatility, and Archetypal Role alignment. Rather than evaluating surface correctness, Mentim focuses on structural integrity across state transitions and cross-context behavior. This document presents the conceptual foundations, formal definitions, and initial measurement specifications of the framework. It is intended for researchers, AI developers, evaluation teams, and governance institutions interested in systematic approaches to AI reliability and interpretability. Version 1.0 represents the initial public release of the Mentim research program. Future revisions may include empirical validation results, calibration studies, and expanded technical appendices.
Mentim (Fri,) studied this question.