Auteur : Vincent Aufray Abstract Paper V of the ÆTHER framework predicted that increasing glucose variability constitutes an early-warning signal of metabolic instability preceding diabetes onset (Prediction S1), based on bifurcation theory and synthetic cohort validation (RR = 1.50). Here, we empirically corroborate Prediction S1 using two complementary approaches: (i) real-world cross-sectional data from NHANES 2017–2020, and (ii) a gold-standard-informed simulation calibrated to continuous glucose monitoring (CGM) literature. In NHANES (n = 3,138 adults without prevalent diabetes), glucose variability was conservatively estimated from the discrepancy between fasting glucose and HbA1c-derived mean glucose. Individuals with estimated CV ≥ 30% exhibited a markedly higher prevalence of prediabetes (HbA1c 5.7–6.4%), yielding a risk ratio RR = 2.36 [95% CI: 2.09, 2.65] and near-perfect discrimination (AUC = 0.945, p < 0.0001). Permutation testing confirmed that this association was not attributable to chance. To calibrate effect size under optimal measurement conditions, we generated a gold-standard-informed CGM simulation (n = 1,000, 14-day monitoring) with direct CV measurement. In this setting, elevated CV remained robustly associated with prediabetes (RR = 1.68 [1.55, 1.80], AUC = 0.739), consistent with longitudinal estimates reported in the literature (HR ≈ 1.5–2.0). Together, these results demonstrate that Prediction S1 survives progressively stricter measurement constraints, exhibiting a coherent effect-size gradient from synthetic proof-of-concept to empirical data and gold-standard-informed calibration. While causality cannot be inferred from cross-sectional or simulated data, this work provides strong empirical corroboration of S1 and motivates longitudinal validation using real CGM cohorts such as the UK Biobank.