The paper presents a methodological investigation into the reuse of digital survey data from the Mausoleum of Theoderic in Ravenna, with the aim of exploring its analytical potential beyond conventional metric documentation. Based on data acquired through laser scanning and photogrammetry, the study examines the geometry of the ashlars forming one of the ground-floor arches, characterised by interlocking “bayonet” joints. An initial vector-based graphical analysis, carried out in view of prototyping and 3D printing operations, raised a question concerning the geometric rules that governed the cutting and assembly of the blocks. In a subsequent phase, the same investigation was verified through a multimodal generative Artificial Intelligence model (LLM/MM), in order to test its ability to reproduce a critical interpretation of the geometric data. Although the AI demonstrated a solid capacity for metric extraction, it failed to identify the underlying geometric rules, thereby confirming the irreplaceable value of human interpretation in recognising the constructive logic of historical architecture.
