How the relationship score is calculated
Every direction begins at zero. Documented conditions activate declared rules, dated events change those conditions, and the same published weights are applied across the atlas.
1. A directional unit of analysis
The model scores the posture of one state toward another from −100 to +100. Reciprocal directions use independent evidence, so they may differ. The current production methodology is frozen: changing a weight, threshold, band or override requires a new version and review.
2. Six visible components
Treaties, war, military support, exercises and force posture.
Formal partnerships, high-level dialogue, expulsions and severed relations.
Trade interdependence, sanctions, tariffs, investment and technology controls.
Shared institutions, joint positions and sustained multilateral coordination.
Territorial claims, coercion, hostile acts and active dispute-management.
People-to-people, education and cultural links; visible for audit but not weighted until global evidence is more comparable.
Societal evidence is shown for audit but currently carries zero weight because globally comparable coverage remains insufficient.
3. Score bands
4. Evidence and time
Baseline conditions are reconstructed for 2015 using rules visible in the evidence ledger. Later events can persist structurally, decay as a pulse or end on a declared date. Annual curves also use directional merchandise trade and UN General Assembly voting alignment where available.
Active interstate war imposes an explicit negative ceiling. A recent lethal interstate clash also limits the maximum score. Overrides are applied after the weighted component calculation and remain visible to the reader.
5. Confidence is not the score
The relationship score describes the calculated posture. Confidence describes evidence coverage. The separate evidence-quality label distinguishes broad sourced evidence, documented bilateral evidence and directions supported only by quantitative indicators.
6. Sensitivity and guardrails
Every rules-based baseline is recalculated under five scenarios: the published weights, security-first weights, geo-economic weights, a de-escalatory impact case and an escalatory impact case. Automated checks reject missing reverse directions, undocumented changes, inconsistent quantitative thresholds and drift from the frozen contract.