The Independent Verification Authority (IVA) is a neutral evaluation layer that audits outputs from multiple large language models and issues standardized reliability judgments. This document defines IVA Methodology v1.0: the five signals (Divergence, Consistency, Hallucination Risk, Omissions, and the Composite score), verdict thresholds, the structure of the IVA Certificate of Audit, alignment with the EU AI Act (Regulation 2024/1689), and known limitations of this initial specification. Version 1.0 is published openly for review and citation. Public critique is invited; substantive contributions informing subsequent versions will be acknowledged in the release notes of subsequent versions.
Maarfia et al. (Fri,) studied this question.