Supply Chain Risk Scoring: How Raw News Becomes Severity-Weighted Geographic Intelligence

The Problem: Unstructured Information, Structured Decisions

Every day, thousands of news articles, government notices, wire reports, and shipping advisories describe events that affect global trade. A refinery explosion in Texas. A port strike in Piraeus. An export ban on rare earth minerals from a major producer. Each of these carries different implications depending on the commodity, the corridor, the duration, and the severity of the disruption.

The challenge is not information scarcity — it's information overload without structure. A commodity trading desk cannot act on a wall of headlines. An insurance underwriter cannot price cargo risk from a PDF digest published weekly. What both need is a pipeline that converts raw, unstructured text into structured, severity-scored, geographically tagged event records — delivered fast enough to matter.

This is the core problem that supply chain risk scoring solves: bridging the gap between what happened and what it means for a specific trade flow, position, or exposure.

From Text to Classified Event: The NLP Pipeline

The first stage of any credible risk scoring system is event detection and classification. Natural language processing (NLP) models ingest source material — news articles, government feeds, maritime advisories — and identify whether a text describes a trade-relevant event. Not every article about a country's economy is a disruption signal. The system must distinguish between background noise and actionable events.

Once detected, events are classified into types: port closures, strikes, sanctions, infrastructure failures, tariff changes, weather disruptions, geopolitical escalations, and more. Each event type carries a different risk profile. A strike at a major container terminal behaves differently from a tariff announcement — different duration curves, different affected commodities, different geographic footprints.

Disruptis processes over 2,400 news sources, wire services, and government feeds daily, running each through classification models that tag event type, affected commodities (across 18+ categories), and geographic coordinates. The result is a structured record, not a headline — machine-readable, queryable, and ready for integration into risk dashboards and trading systems.

Severity Scoring: Why Direction and Magnitude Both Matter

Classification alone is insufficient. A port closure lasting two hours and one lasting two weeks are categorically different risks. A sanctions package targeting a minor export sector and one targeting a country's primary commodity output demand different responses.

This is where severity scoring adds analytical depth. Disruptis uses a bidirectional severity scale from -4.0 to +4.0, where negative values represent disruptions (supply contractions, access restrictions, infrastructure damage) and positive values capture restorations (sanctions relief, port reopenings, capacity expansions). This bidirectional approach matters because risk is not monotonic — the resolution of a disruption is itself actionable intelligence.

Severity assignment considers multiple factors: the scale of the affected trade flow, the replaceability of the disrupted route or source, historical precedent for similar events, and the specificity of the available information. A severity of -3.5 on a Strait of Hormuz transit disruption signals a materially different risk posture than a -1.2 on a minor inland logistics delay.

For trading desks, severity scores translate directly into position sizing and hedging urgency. For underwriters, they inform exposure recalculation. For logistics operators, they determine whether rerouting decisions need to be made now or monitored passively. The methodology behind this scoring process is documented in detail on the Disruptis methodology page.

Geographic Intelligence: Mapping Events to Corridors and Coordinates

The final layer — and the one most often missing from traditional risk reports — is geographic precision. An event tagged to "Middle East" is analytically weak. An event mapped to the Strait of Hormuz, with coordinates, affected vessel classes, and downstream port dependencies, is operationally useful.

Disruptis maps every detected event to geographic coordinates and, where applicable, to specific trade corridors. This enables spatial queries: what is the current aggregate severity load on the Suez Canal corridor? Which West African crude export terminals have experienced disruptions in the past 30 days? Where are restoration events clustered?

This geographic layer transforms risk scoring from a flat list of events into a spatial intelligence surface. Overlaying severity-weighted events on chokepoint and corridor maps reveals concentration risk that headline scanning cannot. It also enables time-series analysis — tracking how severity loads on a given corridor evolve week over week, identifying escalation patterns before they reach critical thresholds.

Operationalising the Output

Structured event data delivered as daily Parquet files — with event type, severity score, commodity tags, coordinates, and corridor mappings — integrates directly into the systems where decisions are made. Trading platforms, risk engines, underwriting models, and logistics planning tools can all consume this data programmatically. Explore the data schema and preview to see the output format.

The value of supply chain risk scoring is not the score itself. It is the compression of thousands of unstructured signals into a structured, severity-weighted, geographically precise intelligence layer that professionals can act on within their existing workflows. That is the pipeline Disruptis delivers daily.