Methodology

1. Open sources only

Every ranking comes from a public source that allows reuse. The source name, a link, the licence and the date we fetched the data appear under each ranking.

2. Automatic collection

Most rankings are filled by small programs that download the numbers from an official data service (for example the World Bank API) or read a published table. Nothing is typed in by hand. The programs use one reference year per ranking, so every row of a list is from the same period.

3. Checks before publishing

• Country names are matched to ISO country codes; anything we cannot match is reported instead of guessed.
• Where a source publishes a world total, we compare it with the sum of the rows.
• A refresh that loses many rows or returns too few is rejected, and the previous version stays online.
• Rankings with data older than 120 days are flagged for review.

4. AI-collected rankings

1 ranking was gathered by an AI research agent that reads web pages when no downloadable dataset exists. These carry an “AI-collected · unverified” label and list every page the numbers were read from. They pass the same format checks, but a person has not yet compared them with the source. Treat them as approximate.

5. Limits

Definitions and methods differ between countries, many figures are estimates, and the latest available year is not the same everywhere. Several rankings (for example AI supercomputers) are census-style counts that miss what is not public, so they are lower bounds.

Sources used