Key points are not available for this paper at this time.
Inconsistent and erroneous file naming is a recurring information management problem in BIM and Common Data Environment (CDE) workflows, affecting document identification, retrieval, traceability, and control. This problem was observed during work on a real-world modular railway station project, where received documentation included files prepared under heterogeneous naming practices. This paper proposes the Level of Naming Compliance (LoNC), a multi-level framework for assessing file name compliance in BIM/CDE environments. The framework evaluates file names as structured information containers and combines three validation stages: syntactic validation using regular expressions, referential validation against project-specific lookup tables, and LLM-supported semantic validation comparing selected decoded file name fields with descriptive document titles. In the current implementation, semantic validation was limited to discipline and document type. The proposed approach was evaluated using a controlled synthetic validation dataset of 500 file-title pairs, developed from inconsistency types observed in the motivating project. The dataset contained predefined examples of syntactic errors, invalid reference values, semantic inconsistencies, and fully compliant cases. Within this controlled evaluation set, the validator showed full agreement with the predefined expected LoNC levels. Because the dataset contained an unequal distribution of compliance classes, the evaluation was supplemented with balanced accuracy and class-specific performance measures. The results indicate that the implemented workflow correctly operationalises the cumulative LoNC framework under the tested conditions. The findings demonstrate that LoNC provides a more diagnostic assessment than binary file name checking by identifying the stage and type of non-compliance. The approach can support CDE managers and document controllers in quality assurance processes, while future work should validate it on larger real-world datasets and extend semantic validation beyond the selected fields.
Pasalski et al. (Mon,) studied this question.