Where the data comes from

Country learning trails

English demonyms come from mledoze/countries and its contributors, under the Open Database License (ODbL-1.0). The derived demonym dataset includes attribution, retrieval date and a source checksum. Common accepted noun forms also follow the English demonym reference.

The civilization entries, historical timelines and additional trivia link to their references beside the text. Historical territory overlaps modern borders; a shared empire does not make those countries interchangeable. Ancient standards and modern reconstructions are labeled separately from national flags. Illustration credits link to the original Commons files.

Religion profiles separate official status from population affiliation. Dated figures come from national censuses in Canada, India, South Africa and Australia, and from Pew Research Center for the United States. The profiles link to the exact reports. Selected categories do not always sum to 100%, and affiliation is not the same as religious practice.

Coverage is still partial. Unchecked status, unknown ancient flags and missing timelines are identified on the country page. Study prompts derived from the older atlas dataset are labeled as imported.

The sources

Why more than one source

Wikidata is the best structured record of the world there is, and it is wrong in specific, predictable ways for the questions a geography quiz wants to ask:

• Borders. Its shares border with counts sea boundaries, which had the United States bordering Japan and Tonga. Every pair is now checked against the real polygons, so only coastlines within about two kilometres of each other count.
• Languages. Speaker counts exist for a handful of countries and nowhere else, and official-language lists are not the same question — Sweden did not list Swedish. The Factbook publishes a breakdown with percentages for almost every country.
• Government. Wikidata lists every form that applies at once: China is a communist state, a socialist state, a unitary state and a one-party state, so the question had four right answers. The Factbook states one type per country.
• Wars. Participants are recorded as the state that fought, which before about 1950 no longer exists — World War II lists Nazi Germany and the Soviet Union, neither of which has a country code. The major wars are written out by hand against their modern successors.
• People. A leader is modelled as the holder of an office, and the office often belongs to a state that has since dissolved, so Muammar Gaddafi appeared nowhere on Libya's page. Offices are looked up both ways round now, and a curated list fills the rest.

What is a thread

Any fact shared between countries gets a page of its own, and every mention of it anywhere on the site links to that page. That covers languages, wars, currencies, religions, organisations, forms of government and highest points, and also climates, landmarks, and every commodity a country buys, sells, mines or makes. The economy fields arrive as bare strings from the Factbook and are wrapped into the same shape as the rest at load time, which is why "crude petroleum" behaves exactly like "Spanish".

What the quiz will and will not claim

Most of this dataset arrives in bulk and carries no source of its own. Where an answer depends on a definition rather than a fact, the definition is written down and shown under the question, so a disagreement is with a stated rule instead of a silent one.

City population

City proper, not metropolitan area, as recorded by GeoNames. Census years differ between countries, so only gaps wider than 15 per cent are asked about.

Continent

Seven-continent model. A country spanning two continents is filed under the one containing its capital; the other is accepted too.

Government type

The single form of government named by the CIA World Factbook.

Highest point

Highest natural elevation inside the country's internationally recognised land territory.

Language share

Most widely spoken language by share of the resident population, first and second language speakers together.

Membership

Full member states only, as the organisation itself lists them.

Trade partner

Share of goods exports by value, CIA World Factbook, most recent year reported for that country.

War participation

Principal belligerents only, named by the modern state that succeeded the power that fought. Absence from the list is not evidence that a country took no part.

One thing is deliberately not asked about at all: who fought in which war. The curated table names principal belligerents, and the imported one lists anybody a source called a participant — funding, basing and reconstruction included. Neither settles whether a country took part, in either direction, so questions built on them marked real participants wrong and treated reconstruction as combat. Dates, the order events happened in, and wars fought between exactly two states are asked about instead, because those the data can actually answer. Every war on a country page says which kind of claim it is.

Corrections checked by hand against a named source are listed on the country pages they belong to, under Checked facts, with the date the source was read. They are reapplied after every dataset rebuild, so a refetch cannot quietly reinstate the old value. Running python scripts/audit_data.py reports how much of the dataset that covers — today, not much, and that number going up is the work.

What is in the dataset today

Rebuilding it

python scripts/fetch_data.py pulls everything structured from Wikidata and Wikipedia, resuming from a checkpoint if it is interrupted. python scripts/enrich_data.py layers the Factbook fields on top and repairs what the first pass gets wrong. The hand-written tables live in learngeo/supplement.py and are merged at load time, so a rebuild cannot wipe them.