Methodology

Data Sources

We aggregate data from multiple authoritative and open-data sources, processing each feed through normalisation and geocoding pipelines.

AI Classification

Incidents are enriched using large language models via a provider-agnostic pipeline with automatic fallback across Groq, Mistral, and OpenRouter.

Clustering

Our event rollup algorithm uses haversine distance (100 km radius) and a 24-hour time window to group related incidents into unified Event records.