From IFC validation to publish/hold decisions: A tool-independent quality-gating framework for railway depots

OpenBIM handovers based on Industry Foundation Classes (IFC) increasingly support asset information management and digital-twin ingestion in rail infrastructure, but owner-side acceptance remains difficult to reproduce across toolchains. This paper develops a tool-independent measurement–policy–diagnostics quality-gating framework for IFC deliverables. The framework defines schema-level measurement functions for spatial executability, semantic robustness and staged information completeness; separates these measurements from owner-configurable policy thresholds; and combines non-compensatory publish/hold decisions with failure-driver and dimension-specific severity diagnostics, including margin-to-threshold measures. The evaluation uses 22 IFC2x3 deliverables from two Spanish hybrid railway depots across existing-condition, projected-design and executed-works states. Under the baseline policy, integrated quality-gate pass rates range from 25.0% to 44.4%. Dominant blocking dimensions vary by state: existing-condition files are mainly constrained by proxy proliferation, projected-design files combine semantic threshold proximity and data-completeness gaps, and executed-works files are mainly constrained by georeferencing-regime and anchor-frame inconsistencies. Robustness analysis shows that failed anchor cases are several orders of magnitude beyond discipline-aware tolerances, whereas semantic outcomes in existing-condition and projected-design files are more policy-dependent under proxy-threshold sweeps. In this corpus, strict field-fill policies are too restrictive as baseline gates and are better used as continuous completeness indicators. The reported pass rates are case evidence rather than statistical estimates for rail IFC handovers. The transferable contribution is the measurement–policy–diagnostics architecture, open implementation, anonymized derived outputs and synthetic benchmark for independent replication and cross-tool comparison.