How the quizzes are made from Wikidata, and what broke

Every question of the WikiCards quizzes comes from a fact stored in Wikidata, the free knowledge base behind Wikipedia, and every answer links back to its Wikidata item. Turning those facts into fair questions took more care than expected. Here is what went wrong, measured on 23 October 2026, and what we changed.

Where the facts come from

A collection script queries the Wikidata SPARQL service and Wikipedia once a month and freezes the result in a file: countries with their capital, continent, currency, official language, population, flag and locator map, rivers, peaks, chemical elements, literary works, paintings, films, Olympic Games, World Cup and championship winners, inventions, people, events and places.

On 23 October 2026 that file held 196 countries, 784 places, 619 people, 1,154 films, 258 literary works, 122 events, 120 rivers, 92 chemical elements, 91 paintings, 50 US states and 460 championship editions.

Every day the pages draw 20 questions from it with the date as random seed: the same set for everyone that day, a new one the next day, and each kind of question rotates through its whole stock before starting again.

Pitfall 1: nearly one famous person in five had no English or French label

Since 2024 Wikidata stores many proper names once, under a label for all languages called mul, instead of repeating them in each language. A query that asks for the English and the French label silently drops those items.

On a sample of the 200 most-read people of our card catalogue, 36 had no English or no French label, among them George Washington, Barack Obama, William Shakespeare and Novak Djokovic. The capital of Alaska, Juneau, had only the mul label, so Alaska vanished from the US states quiz.

The fix is to take the label of the language, then mul, then English.

Pitfall 2: queries that scan a whole class time out

The public SPARQL service stops a query after 60 seconds. Asking for every mountain with its elevation, or every novel with its author, ended in an HTTP 504 error or in a truncated JSON answer, depending on the load of the service.

The fix is to start from a list of known items instead of a whole class: our most-read Wikipedia cards, mapped to their Wikidata identifiers through the Wikipedia API, then queried by batches of 100 identifiers. The file is also written after each kind of fact, so one failure no longer loses the others.

Pitfall 3: the Nile was 2,850 km long

Many items carry several values for the same property. The Nile has a length of 6,650 km with a preferred rank, and also 2,850 km with a normal rank. The Amazon has 6,400 km with a preferred rank and four other values, from 3,690 to 6,992 km.

Taking the median of every value makes the longest river question wrong. We take the values of preferred rank when they exist, drop an item whose remaining values disagree by more than 12 %, and only ask which of four rivers is the longest when the first is at least 15 % longer than the second. The Nile and the Amazon, 4 % apart, are never the two candidates of the same question.

Pitfall 4: true in Wikidata, unfair in a quiz

Some facts are correct in the database and still make a bad question. Pop songs often record a single co-writer as composer: Wikidata gives RedOne as the composer of Poker Face, which he wrote with Lady Gaga. We removed music.

The discoverer property also applies to places: Persepolis would be credited to a Spanish diplomat. Discoveries with coordinates are left out.

Franchises keep their current name for past titles: Super Bowl XI, played in 1977, is won by the Las Vegas Raiders, who were the Oakland Raiders then. American leagues only use editions from 2000.

Wikidata lists Oceania twice, as Oceania and as Insular Oceania, which put Tuvalu on no continent. Both are merged. Capitals whose status is disputed internationally, Israel and Palestine, are never asked and are marked in the list of capitals.

Wrong answers are never a prefix of the right one, so Rugby never stands next to Rugby union.

Pitfall 5: thumbnails at 320 pixels are refused

Flags and locator maps are thumbnails from Wikimedia Commons. A width of 320 pixels returns an HTTP 400 error asking to use the listed thumbnail sizes, while 120, 250, 330 and 500 pixels are served. The quizzes use 330 pixels, and 120 for the flag gallery.

The cleaned fact set is open data (CC0) on GitHub: JSON and CSV files, with the method and its limits.

Measured on 23 October 2026 on the Wikidata SPARQL service, the Wikipedia API and Wikimedia Commons. Facts from Wikidata (CC0). WikiCards is an independent game, not affiliated with Wikipedia or the Wikimedia Foundation.