A CodeSights flagship: 1,786 named storms across all nine basins, built from Wikipedia storm articles because the knowledge graph barely records impact (the probe found deaths on 154 storms, damage on 46, and the 1970 Bhola cyclone's season filed as a Roman arch). Each column comes from where its truth lives: dates, fatalities and damage from storm infoboxes with prose fallbacks; winds and pressure from prose where units are explicit, because infobox wind parameters changed meaning across template generations while prose says 175 mph (280 km/h) unambiguously; Saffir-Simpson category from the lead's own words and only in the two basins that use the scale, so typhoons honestly carry NA rather than a borrowed ranking. The leaderboards read like the record books because the tests require it: Bhola's 500,000 dead is the deadliest and the guard's ceiling, Katrina and Harvey tie at 125,000 USD millions with Ian behind, Ioke's 354 km/h leads the winds under the 360 guard, and there is no Category 6 by test. The seasonality stories are assertions too: Atlantic formations must peak in August-September (188 and 182 rows agree) and the southern-hemisphere basins must flip the calendar. The build's gauntlet is documented in the scraper: Meteorological history sub-articles arrived as duplicate storms, a floods article claimed 250,000 deaths that were actually people left homeless, and Cyclone Nargis carried 150 billion dollars of damage until its bare-number infobox value (15300000000, no dollar sign, no unit) parsed as raw dollars. Duplicate titles merge keeping the row that knows the most; storm names recycle, so rows are unique by storm, basin and season. Sixteen acceptance tests.
1,786 rows in one CSV file, with 14 columns in total: Storm, Storm_ID, Basin, Season_Year, Formed, Dissipated, Duration_Days, Peak_Category, Max_Winds_kmh, Min_Pressure_hPa, Deaths, Damage_USD_Millions, Areas_Affected, Storm_URL.
It is built by programmatically scraping en.wikipedia.org, last pulled on 2026-10-13. 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.
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.