Encoding the Typological Reading of an Ottoman Fountain as Point Classification Rules: A Scan-to-HBIM Case Study Evaluated Against Two Independent Annotations

Heritage building information modelling requires deciding which scanned points form which element. This single-case study encodes the typological reading of Alaca Çeşmesi, an Ottoman square fountain, as readable point-classification rules carried into IFC4 elements with heritage property sets. The rules, calibrated on a complete manual segmentation of the same cloud, were compared with two independent blind annotations. The annotators agreed on 97.0%, the unmodified rules with them on 86.3% and 87.0%. These values measure encoding fidelity on one structure, not performance on unseen buildings. Most of the gap is a class-definition difference between calibration and the reference: whether the niche belongs to an arch or a façade. An exploratory post hoc mask for the annotators’ definition raised agreement to 96.7% and 98.1%; this is not an independent validation. A template filled from 16 picked points reached a mean intersection over the union of 0.726 (rules: 0.764); a co-author blind to the earlier landmark values and scores reached 0.713, differing mainly in the interpretation of the niche opening, represented by two picked points. The classification stage was exchangeable for a Random Forest; both IFC files linked into Revit, showing technical, not semantic, validity. Class definitions should be fixed before calibration and checked against a second reference.