Survival analysis you can taste: 615 candy bars and confections with launch years on 447, discontinuation years on 31, and a Still_Sold flag as the censoring indicator, because Wikidata classifies candy so poorly (the chocolate bar class holds 16 instances) that the roster came from Wikipedia's candy categories instead, 983 candidates pared by a purity gauntlet that threw out a bottled water brand, the companies that rode in on brand categories, and nothing else. The discontinued shelf is pure nostalgia: PB Max died in 1994 at five years old despite 50 million dollars in sales, because, per a former Mars executive, the Mars family disliked peanut butter. The dataset's hardest rule protects the living: a Mars bar discontinued in the US in 2002 but sold worldwide stays Y, so prose-level death requires the article's lead to declare it (is a discontinued brand, was a candy bar), and the regex that finds makers had to learn that a bar produced by the Hershey Company is not itself a company. The requested extras ship at their honest density and no denser: ingredients on 174 rows (infobox first, lead prose fallback, markup rejected), calories on 41 (the largest stated in-window figure, so full bars beat fun-size), revenue on 20 normalized to USD millions with the stated year. Launch dates prefer infoboxes, where they hide inside start-date templates that strip to nothing unless you catch the year first, then Wikidata best-rank, then earliest prose mention, so Kit Kat says 1935 rather than the 1970 a Japanese variant launch once supplied. Built with a 14-test acceptance suite.
615 rows in one CSV file, with 13 columns in total: Candy, Candy_ID, Manufacturer, Country, Launched_Year, Discontinued_Year, Still_Sold, Lifespan_Years, Ingredients, Calories_kcal, Revenue_USD_Millions, Revenue_Year, Candy_URL.
It is built by programmatically scraping en.wikipedia.org, last pulled on 2026-10-12. 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.
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.