The recent availability of off-the-shelf commercial cameras equipped with dual-fisheye lenses, each featuring a hyperhemispherical field of view exceeding 180°, has significantly advanced research addressing a long-standing challenge in Heritage Building Information Modelling (HBIM): the accurate, cost-effective image-based surveying in narrow and spatially constrained environments within Scan-to-BIM workflows. However, advances in digital sensor technology alone do not guarantee the production of accurate HBIM models. Beyond data completeness, HBIM requires high-accuracy modelling with sub-centimetre precision. Achieving such precision through photogrammetric workflows requires rigorous control at every stage of data processing, including camera calibration and 3D reconstruction. However, accuracy analysis can often be overlooked when using highly automated commercial software solutions. To fully exploit the wide field of view of dual-fisheye cameras, in terms of both completeness and final accuracy, an appropriate modelling of hyperhemispherical (HH) lenses using dedicated projection models is essential. This requirement has been a major barrier to the widespread adoption of such cameras in HBIM workflows, given the limited availability of commercial photogrammetric solutions that can properly process dual-fisheye imagery. This paper evaluates the calibration and 3D reconstruction accuracy of a 360° dual-fisheye camera (Ricoh Theta Z1) for surveying a narrow space, using fisheye projection models (equidistant and equisolid-angle) recently implemented in Agisoft Metashape. This represents an important advance in Scan-to-BIM workflows using fisheye imagery, offering a user-friendly, accessible software solution that builds on prior community research that highlighted the importance and benefits of explicitly modelling hyperhemispherical lenses. Thus, this practical study aims to provide insights for the HBIM community by demonstrating how low-cost 360° cameras, when rigorously modelled, can be effectively employed in high-precision Scan-to-HBIM workflows. Results show that the equisolid-angle model achieved higher accuracy, both with 180° FoV images and HH images.