This paper presents the framework of an ongoing semantics knowledge management practice. The practice combines geometrical analysis of point cloud and semantic rules to detect 3D building objects for the purpose of achieving computational correctness and efficiency. We assume that improvement in the form of well organized classification (CLA) and order (ORD) is a solution. Statically related knowledge is classified as definition, partial knowledge and ambiguous knowledge and formally revealed with transitions between CWA (closed world assumption) and OWA (open world assumption). To achieve the correctness, formalization is proposed for the validation of the rules through the mapping from Yes/No (Y/N) to True/False (T/F) and a further refinement of CWA/OWA flow. Dynamically a constructive model of knowledge rules organization is proposed to improve the controllability of previous empirically classified rules. A trade-off is illustrated to enable a quickly inheriting implementation.