When meshes lie formalising tacit knowledge for flaw recognition in cultural heritage photogrammetry

This article presents a methodological framework for the semantic characterisation of surface flaws in photogrammetric 3D models, developed within Cultural Heritage digitisation workflows. Rather than relying on pre-defined classifications or automatic detection, the method captures expert corrections during mesh cleaning and formalises them as traceable, interpretable annotations. Through a retro-projection mechanism, these interventions are mapped back onto the pre-corrected model, producing a semantic segmentation of flaws grounded in situated expertise. Each flawed region is analysed using mesh-derived indicators, such as PCA-based descriptors, edge length, and photogrammetric confidence, to identify recurring patterns of deviation. This process supports the construction of a frequency-based taxonomy of common artefacts, structured across low-, mid-, and high-frequency classes and illustrated with real-world examples. The taxonomy does not aim to provide exhaustive coverage or prescriptive correction strategies but serves as a proof of concept for encoding tacit judgement into a reproducible descriptive system. The approach is applied to a single-operator dataset and is presented as a foundational step toward reproducible flaw annotation and structured paradata integration. While further validation across operators and use cases is required, the method already offers a means to bridge procedural opacity and semantic clarity, particularly in workflows where dissemination assets derive from documentation-oriented models. By aligning mesh geometry, operator intervention, and scalar descriptors, the framework establishes a basis for more transparent assessment, future comparative validation, and potential integration into semi-automated classification systems. It also contributes to the ongoing effort to embed semantic meaning and interpretive traceability directly within 3D assets.