International Relations Atlas — bilateral relations dataset

Version 2026.08.12 · 231 countries · 26,565 country pairs · 28,697 scored claims SHA-256 of data.json: 7e7f1203686c3090f82e4e1704afd6cce84407456f4e3666e27ead89bb9bd0b0 Live application: https://international-relations-atlas.netlify.app What this is A structured record of the relationship between every pair of countries in the world. Each pair carries some combination of: dated historical moments with a signed weight, prose assessment, treaties in force, trade and migration flows, UN General Assembly voting alignment, diplomatic representation, geography and colonial history, and forced-displacement corridors. Most fields carry a source URL at the level of the individual claim rather than the record. A single score between −100 and +100 is derived per pair. It is an editorial model, not a measurement, and the constants behind it are published in the application's methodology section. Coverage, stated plainly PairsShare of 26,565 Any record at all22,89986.2% Geography and colonial history22,57885.0% Written record (dated claims, prose or timeline)4,55017.1% Agreements in force2,4399.2% Forced-displacement trajectories1,1074.2% The gap between the first row and the third is the honest shape of this dataset: most pairs hold complete, cited structural facts, and a minority hold a researched narrative. A pair with no land border, no shared language and no colonial tie has a *record of absences*, which is a real description of a real relationship — not a gap. Known limitations These are measured, not estimated, and anyone using this data should read them before quoting a figure. Sampled factual error rate: 6.7% (95% CI 3.4–12.6%), from a seeded random sample of 120 scored claims hand-checked against the historical record in August 2026. Eight contained a confirmed factual error. **Sourcing of scored claims: 3.4% first-hand, 67.3% encyclopedic, 29.3% unsourced.** Roughly two-thirds of citations lead to an encyclopedia rather than to a primary source, and just under a third of claims carry no followable citation at all. **10,374 pairs (39%) have a score resting on UN General Assembly voting alignment alone.** The application withholds the numeric headline for these and shows only a band word. Treat the underlying number as a proxy, not as evidence about the relationship. 77 errors have been found, independently verified twice, and removed. That count is a floor on what was wrong, not a ceiling on what remains. Machine-checkable defects: 0 — no JSON fragments, control characters, impossible dates or self-contradicting accession dates survive in any scored claim. This says nothing about historical judgement, which is where the remaining errors live. Coverage is uneven by region. The written record is strongest for Europe and the major powers and thinnest for small states and the Global South. Anyone computing an aggregate across all pairs is aggregating that unevenness. Treat every entry as a lead to verify rather than a fact to quote, and weight uncited claims lower still. Principal sources Correlates of War (formal alliances v4.1, direct contiguity v3.2, diplomatic exchange 2006 v1); UN Comtrade; UN General Assembly voting records; UN DESA international migrant stock; UNHCR refugee population statistics; UNCTAD Investment Policy Hub; the United Nations Treaty Collection; WTO Regional Trade Agreements Information System; CEPII GeoDist (Mayer & Zignago 2011); OECD DAC; World Bank World Development Indicators; CIA World Factbook; national foreign ministries; and Wikipedia. Every individual claim that carries a citation names its own source in the data. Files data.json — the complete corpus. The SHA-256 above is of these exact bytes, so a reader can confirm they are reading the version cited. README.md — this file. Citing International Relations Atlas. (2026). Bilateral relations dataset (Version 2026.08.12) [Data set]. Zenodo. Cite the underlying source directly wherever one is listed in the record. This reference is for the synthesis, not for facts attributable at first hand. Licence Creative Commons Attribution 4.0 International (CC BY 4.0). Third-party source data retains the licence of its original publisher.