From Model Performance to Screening Readiness: An Audit-Grade Evidence Mapping Study in Prediabetes

Background: Prediabetes represents an intermediate metabolic state preceding type 2 diabetes, where delayed identification can postpone effective, low-cost preventive measures. Methods: We propose a taxonomy-first, audit-grade evidence mapping framework and score each model unit across validation maturity (V), reporting completeness (R), calibration/threshold readiness (C/T), and reproducibility (Re), using conservative coding and dual independent extraction. Results: In a pilot Run V0, we observed recurring gaps in calibration, threshold definition, and modality descriptions that limit decision readiness despite reported discrimination metrics. Conclusions: The framework and reproducible workflow provide a transparent method to quantify reporting and decision-readiness gaps and are designed to scale to a larger Run V1 under a frozen codebook.