How we collect, score, and present crisis intelligence
RedAlertDirect aggregates crisis data from the following verified public sources:
All sources are publicly available and free to access. We do not rely on anonymous or unverified sources.
Raw event data is processed using Anthropic's Claude AI model to generate structured summaries, timelines, and context. The AI is instructed to:
All AI-generated content is reviewed and labelled. We acknowledge AI systems can make errors — if you spot an inaccuracy, please contact us.
Each crisis is assigned a severity score based on a weighted composite of the following factors:
| Factor | Weight | Data Source |
|---|---|---|
| Displacement scale (IDPs + refugees) | 30% | UNHCR, OCHA |
| Civilian casualty rate | 25% | ACLED, news aggregation |
| Escalation trajectory (30-day trend) | 20% | GDELT, ACLED |
| Humanitarian aid access | 15% | ReliefWeb, OCHA |
| Regional spillover risk | 10% | AI assessment |
Scores are recalculated with each data refresh. A score of 90+ indicates a critical humanitarian emergency.
Crisis summaries are updated daily via automated data ingestion. Breaking developments may trigger more frequent updates. The "last updated" timestamp on each crisis card reflects the most recent data refresh, not necessarily a change in the situation.
Crisis data is inherently incomplete. Access restrictions, reporting delays, and political sensitivities mean some situations are underreported. Severity scores are models, not ground truth. Always cross-reference with official sources before making decisions.