Dataset for BIM-Context-Aware Evidence-Traceable Safety Review Using Accident Narratives

This dataset provides de-identified structured accident-narrative data and supporting prototype materials for BIM-context-aware, evidence-traceable safety review. The dataset was derived from publicly available OSHA Severe Injury Report narratives and was processed to convert unstructured accident descriptions into structured safety fields that can support BIM-based safety reasoning, evidence retrieval, and safety advisory generation. The shared package includes: (1) a de-identified master structured accident dataset; (2) a de-identified researcher-reviewed validation subset; (3) a BIM object risk profile; (4) an internal validation summary report; (5) prototype examples showing BIM context extraction, augmented evidence packaging, and generated safety advisory output; and (6) cleaned workflow notebooks for label validation, risk-profile generation, and context-aware evidence generation. The core structured fields include BIM-related object classes, construction activity labels, work-phase information, hazard triggers, injury outcomes, OSHA reference identifiers, and de-identified accident narratives. Direct identifying fields, including employer names, address fields, geographic coordinates, inspection identifiers, and internal personal paths, were removed before publication. The example BIM context files use a generic sample structural model, masked element identifiers, and masked X/Y coordinates. The materials are intended to support transparent reuse of the structured dataset, validation subset, and prototype workflow. The dataset can be used for safety knowledge structuring, BIM-linked risk review, accident evidence retrieval, safety decision support research, and reproducibility checks for the accompanying study. The generated safety advisory example is provided only as a prototype output and should not be interpreted as a certified safety plan or regulatory compliance document.