How the country risk score is calculated

Source indicator. A publisher measures a particular concept using its own scale and observation period. For example, the World Bank Worldwide Governance Indicators are perceptions-based governance measures; the INFORM Risk Index concerns humanitarian crisis risk. Their values and directions are not interchangeable.

Salient One risk band and score. We apply published thresholds to available indicators, then combine rated groups using the calculation below. A band is our interpretation of the selected inputs, not a classification issued by the underlying publisher. Review the machine-readable method alongside the source and licence directory.

Observation date and coverage. Source editions differ. A recent profile computation does not make an older observation current, and a missing indicator is not a low-risk observation. Country profiles show source dates and coverage; compare those before interpreting differences between countries.

Country context and entity evidence. A jurisdiction-level score describes selected national conditions, not a legal finding about an organisation or transaction. Verify identity, ownership, sector and current official records separately.

The sourcing dataset separates upstream facts from Salient One derivations. Each row can identify the source directory URL, a record-level URL, source effective date, retrieval time, derivation owner, method version, assessment basis and enforcement status. Retrieval time is operational freshness; it is never treated as the date the source fact became effective.

World Bank forest-area trend. Salient One reads indicator AG.LND.FRST.ZS, compares the oldest and newest available values in the retrieval window and calculates the relative annual change against the oldest forest-area percentage. Positive loss of at least 0.2%, 0.7% and 1.5% per year routes to moderate, high and severe review bands. This is a Salient One calculation from World Bank forest-area data, not a Global Forest Watch tree-cover-loss measure.

EUDR evidence-review scope. Existing country × covered-commodity combinations are published as Salient One screening flags with no public severity band. They help route evidence collection; they are not the European Union country benchmark. The official country classification remains a distinct legal source and must be checked at decision time.

Conflict-area mineral screening. The current country × mineral assessment is a fixed Salient One review-priority snapshot based on a 2021–2024 evidence review. Its published bands are Salient One screening bands, not OECD designations. The European Commission describes the official CAHRA list as indicative, non-exhaustive and periodically updated, so current primary material remains mandatory.

CBP enforcement actions. Only actions whose current source status is Active or partially modified/active contribute to current evidence. Inactive actions are excluded. A WRO routes to moderate because CBP describes its evidence as reasonable but inconclusive; a Finding routes to high because it is a determination. The source status and effective date remain visible.

Product safety and labour. Source observations are mapped to the common Salient One sector taxonomy. The labour-goods adapter accepts only an official publication with a verified stated effective date, at least 80 resolved countries and at least 150 unique goods. Goods without a high-confidence sector rule remain in the source table but are not assigned to a sector. Any qualitative routing band calculated by Salient One is identified through derived_by and method_version. Absence of a subject match is a coverage statement, not evidence of low risk.

Download the machine-readable method (JSON)

Each country is evaluated across seven risk groups containing the available country indicators. Eligible indicators are assigned a risk level — low, moderate, high, or severe — based on fixed, transparent thresholds derived from the source's own scale or established risk-assessment practice.

Within each group, the most elevated available indicator determines the group band. The composite then averages the seven group scores, excluding groups without data from the denominator. This prevents a cluster of correlated indicators from receiving disproportionate weight.

Labour and human rights uses World Bank WGI Voice and Accountability, child-labour outcomes and listed labour-risk goods. Voice and Accountability is a broad perceptions-based measure, not a press-freedom index or a complete measure of human rights. It is excluded from the five-dimension institutional governance mean and used once in this group. ILO ratification is context only and does not establish enforcement; the absence of listed goods does not establish low risk. Restricted legacy indicators are withheld from new releases.

The composite score is reproducible from the group results. Low, moderate, high and severe groups score 0, 1, 2 and 3. Salient One sums the available group scores, divides by the number of groups with data, computes a base score as (average ÷ 3) × 100, then publishes min(100, round(base score × (100 ÷ 65))). The calibration preserves the engine's established 0–100 display scale; the qualitative band is determined from the uncalibrated average: low at ≤0.4, moderate at ≤0.9, high at ≤1.6 and severe above 1.6. A high group prevents a low overall band, and a severe group sets a minimum overall band of high.

A minimum of two indicator groups with data is required to compute a composite rating. A group without sufficient evidence for a rating is excluded from the denominator rather than treated as low risk. Group coverage does not imply every underlying indicator is available.

Directive (EU) 2024/1760, as amended, provides a corporate sustainability due diligence framework. Salient One country indicators can help identify geographic questions for further assessment. They do not establish adverse impacts, the adequacy of company measures or compliance with an obligation.

This indicative mapping connects the seven risk groups with selected provisions. Every contribution is partial country context; company, project, stakeholder and transaction evidence remains necessary. Consult the current consolidated directive and applicable national law for scope and timing.

The signal engine compares a user-supplied name against records from commercially permitted public-list sources, subject to the source availability shown in the coverage directory. UN and multilateral development bank lists are not fully searched; no match does not establish clearance. It also matches against listed aliases and known spelling variants to catch alternative renderings of sanctioned names.

Matching uses a weighted ensemble of four complementary techniques, each contributing to a combined confidence score (0–1):

• Jaccard token overlap — measures how many words (tokens) the query and the target name share, normalised by the total number of unique tokens. Robust to word reordering and partial name matches.
• Dice coefficient on character bigrams — compares overlapping two-character sequences between the query and target. Effective at catching typographical errors and minor spelling variations.
• Levenshtein edit distance — counts the minimum number of single-character insertions, deletions, or substitutions needed to transform one string into the other. Normalised by string length to produce a 0–1 score.
• Double Metaphone phonetic hashing — encodes names into pronunciation-based keys, so names that sound similar (e.g. Muhammad / Mohamed) hash to the same or nearby values regardless of spelling differences.

The four scores are combined into a weighted ensemble, with higher weight given to token-level matching for multi-word names and phonetic matching for single-word names. A configurable minimum confidence threshold (default 35%) filters results before they are returned.

Transliteration support covers Cyrillic (Russian, Ukrainian, Serbian), Arabic, and Chinese (Simplified) scripts. Names in these scripts are automatically Romanised before comparison, enabling cross-script matching without requiring the user to know the original script.

When optional fields are provided — date of birth, passport or ID number, IMO number, address, or place of birth — the engine applies exact and near-exact matching on those fields as a secondary signal. A field-level match boosts the overall confidence score but is not required for a result to be returned.