KinaviAccident-8293

KinaviAccident-8293 is a large-scale, structured global maritime accident dataset compiled from official investigation reports issued by 89 countries/regions, covering 8,293 accidents from 1912–2025. It provides 63 standardized fields per record, combining core metadata (date, location, vessel and accident type), casualty and loss outcomes, and rich narrative annotations such as brief/full summaries, lessons learned, recommendations, and detailed causal content (root causes, human/technical/environmental/organizational factors, causal chains, and timelines). The release includes a full multi-modal pipeline from original PDFs to parsed Markdown and structured JSON/CSV, along with Python scripts for statistics, quality assessment, and visualization; reported data quality is Grade A overall (91.8%) with high completeness for critical fields. Designed for statistical analysis, risk assessment, safety research, knowledge-graph construction, and ML/LLM applications (including SFT-ready formatting), the derived dataset is licensed under CC BY-NC 4.0, with ongoing updates noted through June 2025 (last updated 2026-02-03).