Reported Tunnel Failures and Severe Losses: An Evidence-Graded Global Database and Forensic Atlas (Corpus Release 3)
Description
This researcher-curated secondary dataset contains 113 reported and coded tunnel failures, severe losses, and high-value near misses from 48 countries or cross-border settings. Candidate incidents were identified from publicly accessible official investigations, regulator and owner records, peer-reviewed and technical publications, professional case histories, and contextual sources. Each incident is linked to source metadata, an evidence grade, field-level uncertainty records, causal-chain coding, and chronology information. The evidence distribution is 10 Grade A, 46 Grade B, and 57 Grade C records; the primary A/B analytical subset contains 56 records. The deposit contains harmonized CSV tables, controlled vocabularies, validation outputs, and manuscript-aligned figure-source data. It is a corpus of reported and coded cases and is not an exposure-normalized census of global tunnel-failure rates. Full third-party reports, journal articles, photographs, drawings, screenshots, and substantial text extracts are not redistributed; source citations and access links are provided in data/sources_register.csv. AI-assisted web-search tools were used only to identify candidate sources; dataset inclusion, extraction, coding, grading, and final validation were performed by the authors against the cited materials.
Files
Steps to reproduce
Unzip the package. Review README.md and docs/data_dictionary.csv. Run `python scripts/validate_release.py` from the package root. Primary analysis should use data/incidents_master.csv filtered to analysis_subset = primary_AB; contextual C-grade records should be analyzed separately. Figure-source CSVs are provided in figure_source_data/. Missing numeric values must not be interpreted as zero.
Institutions
- National Institute of Technology, Asahikawa CollegeHokkaido, Asahikawa
- Akita UniversityAkita, Akita
- Kanazawa UniversityIshikawa, Kanazawa