While the geometric precision of building models is critical in microclimate computational fluid dynamics (CFD) simulations, the reliance on two-dimensional (2D) facility management (FM) drawings, which often lack exterior finish details, remains a prevalent practice. This study quantitatively analyzed the geometric deviation between 2D FM drawings and three-dimensional (3D) light detection and ranging (LiDAR) scan data of an aging large-scale building to examine the inherent uncertainties of both datasets. Using a LiDAR-based Scan-to-Plan approach, the as-is 2D exterior contour was extracted, and a Python-based centroid auto-alignment with a KD-Tree algorithm was applied to mathematically quantify deviations based on 15,224 precisely matched points. The measurement, registration, and extraction uncertainties of the scanning pipeline were further incorporated into an error budget to distinguish scan-induced error from actual form differences. The analysis revealed a mean bias error (MBE) of +0.1160m, empirically demonstrating a systematic outward offset caused by the omission of exterior finishing materials, while the 95th percentile error (P95) reached +0.6130m and only 13.91% of the data points fell within the acceptable construction tolerance (±30mm). The error budget confirmed that the observed deviation predominantly reflected real geometric differences–omitted finish thickness and post-construction deformation–rather than scan noise or human error. By cross-examining the volume underestimation of drawing-based models and morphological overrepresentation of LiDAR scan data, this study highlights the necessity of recognizing geometric gaps and carefully calibrating the data