<p> This article proposes a hybrid geometric modeling approach for the reconstruction of archaeological objects based on historical sources and fragmentary remains. The case study focuses on the Shakhriston Hill complex located in the city of Kuva, Fergana Region, where diverse object types are examined, including residential traces, elements of temple and funerary structures, storage pits, and ceramic artifacts. The workflow is organized into three stages. First, point clouds acquired through photogrammetry and terrestrial laser scanning (TLS) are cleaned, registered into a unified coordinate system, and semantically segmented. Second, a structural “skeleton” of the object is reconstructed through parametric fitting of elementary geometric primitives (planes, cylinders, cones, spherical or paraboloidal segments, and surfaces of revolution). Third, local gaps are completed using spline-based smoothing, while historical proportions, modular constraints, and symmetry rules are incorporated into the optimization objective as regularization and penalization terms. Source reliability is accounted for through weighting, and the final model is evaluated using a combined set of indicators, including geometric accuracy, historical consistency, and evidential coverage. The paper emphasizes a two-step reconstruction logic – from an initial visual hypothesis to an accuracy-controlled practical reconstruction – which enables early reduction of alternative solutions and saves time and financial resources. The results support the generation of verifiable visual materials for researchers, including semantic relationships between elements and uncertainty maps. The proposed hybrid model enhances stability in cases where physical fragments are incomplete by formally embedding historical constraints into the reconstruction process, thereby providing a methodological foundation for restoration and scientific analysis. To ensure reproducibility in an open-source environment, the fitting and optimization stages can be implemented using R packages, including robust regression, nonlinear fitting, and smoothing functions for stable parameter estimation. In addition, linking the reconstruction to paradata is recommended to document which elements are based on measured evidence and which rely on historical inference.</p>