Methodology

How to read our figures

Every figure names its source, the year it covers and the date we last refreshed it. Figures that are not final, reported actuals carry a tag with its own symbol and word: Estimate, Projection, Preliminary or Not yet reported. Colour is never the only sign.

Rankings compare countries on one common year by default, so a country is never ranked against another’s older figure. “Latest available” uses each country’s most recent year instead, up to two years earlier; the table then marks the older years and says how many there are.

A rank says where a country falls on one measure, largest first. It is not a judgement, and “Higher” on Compare pages only marks the larger number.

Import reliance

Import reliance is the share of what a country uses that it buys from abroad, after taking off what it sells. For potash and phosphate it is the share of supply instead.

Term	How it is worked out
Import reliance	(imports − exports) ÷ use, or ÷ supply for potash and phosphate
Use	oil and gas: consumption, as EIA reports it; crops: production + imports − exports
Supply	potash and phosphate: production + imports − exports, that is production plus net imports

FAOSTAT publishes fertiliser use, but only agricultural use: what farmers apply. Its trade figures cover every use, industry included, and goods passing through on their way elsewhere. Set against farm use alone, net imports came to more than all of a country’s use in dozens of countries, the United States and France among them. Supply puts trade against figures of the same scope. FAOSTAT estimates most fertiliser trade figures itself; a row that uses one carries the Estimate label.

A negative value means the country sells more than it buys. We then show exports as a share of production instead (“Exports 60% of what it produces”).

Import reliance	Label
50% or more	Imports most
15% to under 50%	Imports some
−15% to under 15%	Low net trade
Below −15%	Exporter

A country that uses none of a resource, by its source, gets no share and the table says so.

When net imports are more than 10% above recorded use, or exports are above production plus imports, we show no share: the row says “Imports exceed recorded use” or “Exports exceed recorded supply” and is left out of rankings and sorting by share and of a country page’s lead imports. A row whose exports exceed supply is still labelled Exporter, with its buyers. Goods passing through, or changes in stocks, can cause this; the data do not show which. In FAOSTAT’s figures, Morocco’s phosphate exports in 2024 were 1.7 times its production plus imports. For crops, exports can also come from an earlier harvest: Ukraine shipped 29.6 million tonnes of maize in 2024 from 26.9 million tonnes grown and imported. Such a crop has no use figure for that year on the maps. A balance more than 10 years older than the latest any country has for that resource is shown with its year and “No recent data”, and is left out of rankings and sorting and of a country page’s lead imports. For potash, where FAOSTAT’s production figure is its own estimate and the USGS Mineral Commodity Summaries give the same year’s mine production, in potassium oxide (K2O), more than twice or less than half of it, we use USGS’s production everywhere that figure appears (the balance, the potash maps, the world total and rankings) and say so: in 2024, FAOSTAT estimated China’s output at 1.1 million tonnes and USGS at 6.3 million tonnes. A change from a year with FAOSTAT’s figure to one with USGS’s is not shown.

The Dependencies tool splits these further, to show where a country sits: imports 75% or more, 50% to under 75%, 15% to under 50%, low net trade (−15% to under 15%), and, for exporters, exports of up to 50% or over 50% of what it produces.

Supplier concentration

For arms and chips the sources give trade but not use, so we cannot say how much of what a country uses comes from abroad. These rows show the main supplier and its share of the country’s imports instead, and the dependence column says “Supplier concentration only”.

For food, fuel and fertiliser, the Dependencies tool also shows the main supplier’s share of a country’s imports, from the trade flows (FAOSTAT for crops, UN Comtrade for oil, gas and fertilisers). Their year can differ from the year of the import share.

We highlight a country and resource when net imports cover at least half of use (of supply, for fertilisers) and one supplier provides at least half of those imports.

Five-year arms totals

SIPRI measures arms deliveries in trend-indicator values (TIV): SIPRI’s measure of military capability delivered, not price or number of weapons. Yearly figures jump with single large orders, and SIPRI advises using them to show trends over periods of time, so we show totals over five years.

SIPRI’s terms allow reuse of under 10% of its dataset, so we publish these derived totals only and offer no download of the underlying records.

Supply chains

Each supply chain page shows one flow for one period: the latest year its source reports (for arms, the latest five-year total). We never mix years in one diagram, and flows from different sources can have different latest years, so each page says which year it shows. For UN Comtrade’s flows, a new year is used once the countries that have reported it cover at least 85% of the year before’s flows; until then the page shows the year before, as “the latest year most countries have reported”.

The flows are the main bilateral flows each source publishes, not all trade. FAOSTAT’s detailed trade matrix is cut to each importer’s main suppliers; UN Comtrade’s flows are as importers report them; UNHCR lists smaller origin and host pairs together. Totals on these pages are totals of the flows shown.

We use the importer’s report because customs records of what comes in are usually the more complete side of a trade. When an importer’s quantity cannot be used, we use the exporter’s report of the same flow instead, and the page names those flows (“United States → France, as the exporter reports it”). That happens when the quantity cannot be in kilograms (its price per kilogram is under a tenth of the product’s that year, such as litres entered as kilograms), or, for gas, when the weight is UN Comtrade’s own estimate and the importer gives no volume of its own. Where a country reports gas by volume only, we use its cubic metres. For oil and gas we also use the exporter’s report of a flow the importer does not report at all: Germany names no partner for its pipeline gas, so Norway → Germany is Norway’s figure. Exports to UN Comtrade’s “Other Asia, not elsewhere specified” are not taken for Taiwan there, as exporters such as Saudi Arabia use it for buyers they do not name. If neither side’s figure can be used, the flow is left out.

Crude oil flows into the United States are EIA’s US crude oil imports by country of origin, in barrels a day; the other crude flows are UN Comtrade’s. Converted from tonnes at an average 7.33 barrels a tonne; heavy crudes (Canada, Venezuela) hold fewer barrels a tonne, so their flows read up to about 15% high.

Concentration is the Herfindahl–Hirschman index on suppliers’ shares of the flows shown: the sum of the squared percentage shares, from near 0 (many small suppliers) to 10,000 (one supplier). As in the 2023 US merger guidelines, 1,000 to 1,800 counts as moderately concentrated and above 1,800 as highly concentrated. For a country, it is worked out on its own suppliers. Refugees by origin and host country have no concentration: countries of origin are not suppliers. Those figures are people hosted at the end of each year, not people moving that year.

Common years, preliminary years and coverage

Sources publish in stages: a few countries report a year early, most report later. The common year on each indicator page is the latest year that at least 80% as many countries have reported as in the best of the five years before it. Later years that some countries have already reported are preliminary; the page lists them and never uses them as the default until coverage catches up.

Every ranking says how many countries and territories it covers, for example “142 of 217 countries and territories reported for 2025”. Pages for measures that fewer than 20 places report are published but not offered to search engines.

Countries, territories and borders

We show data for the unit the source reports and never move figures from one place to another. Where a source includes a territory in its parent’s figures, or reports a disputed area separately, the country page says so in its coverage notes. Maps use Natural Earth’s 1:50m boundaries, which follow the de facto line of control; boundaries shown do not imply endorsement.

Summaries and sentences

The sentences at the top of country, Compare and indicator pages are written by templates from the figures on the page, not by AI. Each sentence names its source and year, and a sentence whose figure is missing is left out rather than guessed.

Some texts are drafted with AI and approved by the editor. The next section says how.

AI drafting

These texts are drafted with AI: the summaries on the United States, China and Russia dossier pages, the notes on the membership history pages and the summaries on the disputed river pages, and the summary of each closed week on This week. The first drafts of About, Privacy, Methodology and the introductions to the five topic sections were made the same way. Dossier, history and river texts carry the label “Drafted with AI from cited sources and reviewed by the editor”, which links here.

Facts first. Each text has a facts file kept with the site’s code. Every fact in it has its value, unit and period, the name of its source, a link to it and a date. A dossier’s facts are the figures its page already shows; a history’s are the dated joins and departures of the grouping, each with a citation such as a treaty or an official announcement; a river’s are its treaties, dam milestones and official statements, each dated and cited. A fact that has not yet been checked against its source is marked as unverified, and nothing is drafted from a file that has one.

Drafting. A script sends a text’s facts file to Anthropic’s Claude Sonnet 5.5 model (claude-sonnet-5-5) with a fixed prompt, kept with the code in src/data/ai/prompt.md. The prompt asks for plain, neutral text written only from its facts: no forecasts, no blame, no claims its facts do not support, no judging adjectives, and every figure with its unit and period, written with thousands separators (5,150). Countries keep the site’s own names (Türkiye, not Turkey), and a country that joins a second time “rejoined”. The text is about its subject: it never mentions its facts file, tables or itself, and never says what its facts leave out.

Checking. A second script, which is not AI, reads every number, date and country name in the draft and looks for it in its facts file. A figure passes only if it is a fact’s value, written exactly or rounded the way it is written to at least two significant figures, with the fact’s unit and period in the same sentence. Dates must be dates in its facts, and countries must be countries its facts name, under any of their usual names. A draft that talks about its facts file or itself, rather than its subject, fails too. A draft that fails is sent back once with the list of what failed; if the second draft fails too, it is dropped and nothing is published.

Approval. Nothing drafted is published until the editor has read and approved it. Each approved text is stored with the date of approval and a fingerprint of the facts it uses. When one of those facts changes, for example when a source publishes a new year, the text is no longer shown until it has been redrafted and approved again. Texts that are not approved show nothing at all.

The editor’s own words. The editor can replace any draft with text written by hand. Such a text is marked as hand-written, is never overwritten by the drafting script and carries no AI label.

The summaries at the top of other country pages, and every sentence on Compare and indicator pages, are written by templates, not by AI.

Known limits

• Sources revise their history. We keep each release, so a figure you cited can be traced, but it may differ from today’s.
• Trade is recorded by country of origin or of consignment, depending on the source. A supplier can be a trading hub rather than the producer.
• Some countries publish little or nothing. Their pages say which sources cover them and which do not, rather than filling the gaps.
• Estimates carry uncertainty. Figures for countries at war, or with large off-budget spending, are the source’s estimates and carry the most.