When Meshes lie: Tracing Flaws and Extracting Knowledge from Expert Intervention in CH Mesh Processing

Surface flaws in photogrammetric 3D models of Cultural Heritage objects are often corrected manually, yet the rationale behind these interventions typically remains undocumented. This work presents a method to formalise such corrections as semantically meaningful annotations. By segmenting operator interventions during mesh editing and retroprojecting them onto the uncorrected model, the framework transforms both stages into complementary carriers of interpretive paradata. The result is a dual-layer system that enables the emergence of a reproducible flaw taxonomy grounded in expert judgement and supported by geometric indicators such as curvature, edge length, and confidence. This transitional architecture supports both retrospective analysis and future integration into structured validation workflows.