<p>Offline aridity and drought diagnostics typically project widespread terrestrial drying under climate change, whereas fully coupled Earth system models (ESMs) often simulate modest or regionally heterogeneous changes—and in some regions increasing—runoff. This long-standing divergence has been attributed largely to missing vegetation physiological effects and the neglect of sub-annual climate variability in offline diagnostic frameworks. Here, we show that a more fundamental issue is the violation of the diagnostic framework’s structural requirement that potential evapotranspiration (PET) and precipitation (P) act as independent climatic constraints. Using Penman and Penman–Monteith formulations, each with and without thermodynamic deflation via the complementary evaporation principle (CEP), we demonstrate that land–atmosphere feedbacks embedded in conventional PET estimates induce strong negative P–PET correlations (−0.45 ± 0.29; mean ± standard deviation) across land surfaces, which collapse toward near zero (−0.02 ± 0.42) after CEP deflation. Preserving PET–P independence substantially reduces inflation of the aridity index and brings offline diagnostic ET trends closer to ESM projections under a strong-emission scenario (from +0.61 to +0.39 mm yr⁻²; ESM mean: +0.28 mm yr⁻²). These results indicate that structural inconsistencies—rather than missing physiological processes alone—play a central role in the mismatch between offline diagnostics and ESM hydrology. Ensuring that PET is not inflated by land–atmosphere feedbacks is therefore essential for theoretically valid offline hydrologic assessments under a warming climate.</p>