Geometric accuracy assessment is a fundamental prerequisite for the application of high-resolution optical satellite imagery, yet it remains particularly challenging in the absence of ground control points. This study proposes a consistency-driven accuracy assessment framework for ZY-3 satellite imagery that systematically bridges the conventional separation between block adjustment and accuracy evaluation. The framework is built upon a unified geometric error model that accommodates both inter-image consistency assessment and absolute accuracy evaluation within a common mathematical formulation. Recognizing the inherent uncertainties of publicly available reference data, a hierarchical validation chain is established: the Google Earth (GE) and Shuttle Radar Topography Mission (SRTM) datasets are first validated against WorldView imagery to bound their intrinsic errors, and subsequently employed as references for ZY-3 accuracy assessment. This design enables the simultaneous evaluation of planimetric accuracy, vertical accuracy, and inter-image consistency without reliance on ground control points. Experimental validation conducted on 1368 ZY-3 scenes covering a study area of 1.9 million km² yields a planimetric RMSE of 3.33 m and a vertical RMSE of 4.27 m. The results demonstrate that higher inter-image consistency empirically correlates with improved absolute positioning accuracy, and that the proposed framework not only reliably evaluates geometric quality but also provides diagnostic insights that can inform proactive adjustment strategies. The proposed method thus offers a practical and transparent solution for the geometric accuracy detection and enhancement of large-area ZY-3 satellite imagery.