Two joinable datasets on every dated fatal-aviation event Wikidata knows, with death tolls completed from Wikipedia's own infoboxes: 3,035 events spanning 1902 to the present, and a year-by-year rollup built for the most famous chart in aviation safety, deaths falling for decades while flying multiplied. Wikidata models disaster four ways and the roster catches all of them: the aviation-accident and aircraft-crash class trees, the deliberate acts under their own classes (Pan Am 103 a bombing, Iran Air 655 a shootdown, the 9/11 flights hijackings), and flight-modeled events like MH17 whose only tell is a significant-event pointer, each labeled in a Category column rather than hidden. Death tolls obey the graph's own grammar: an unqualified deaths statement is the total, qualified parts sum (Lockerbie ships as 270: 243 passengers, 16 crew, 11 on the ground), and umbrella events stand down for their component flights so 2001 never double-counts. Where the graph is silent the article's aircraft-occurrence infobox speaks: 269 tolls plus aircraft types, operators, occupants and survivors backfilled, every toll tagged with its Deaths_Source. Rows carry date, flight number, operator, aircraft type, origin and destination, country and coordinates, occupants, injured and survivors; the years file totals each year and names its deadliest accident. The deadliest single accident is still Tenerife 1977, at 583. Built with a 20-test acceptance suite anchored on Tenerife, JAL 123, KAL 007's summed parts, the Hudson ditching and the no-double-count rule.
3,149 rows across 2 joinable files, with 26 columns in total: Accident, Accident_ID, Date, Year, Category, Flight_Number, Operator, Aircraft_Type, Origin, Destination, Country, Location, Latitude, Longitude, Occupants, Deaths, Deaths_Source, Injured, Survivors, Accident_URL, Year, Accidents, With_Death_Toll, Total_Deaths, Deadliest_Accident, Deadliest_Deaths.
It is built by programmatically scraping wikidata.org, last pulled on 2026-09-12. Datasets are versioned; older versions stay downloadable.
Yes. Every CodeSights dataset is completely free as a CSV download; a free account is all it takes. Anyone can preview the data without signing in.
Yes. The exact Python scraper that built it is viewable on the dataset page by any signed-in member, so every number is reproducible.
Join the files on Year (deaths_by_year.Year = accidents.Year); the years file is a per-year rollup of the accidents file.
Automated scraping leaves room for error and the underlying sources change over time, so no version is guaranteed accurate or complete. If a number matters, verify it against the original source.