Frequently Asked Questions

A real-time visualization of global instability. Each country gets a 0–100 pressure score based on aggregated incident signals (conflicts, disasters, protests, cyber attacks, etc.). The map uses color bands to show which regions are under stress.

The score represents "pressure" — an abstract measure of instability signals. It's not a prediction or safety index. Higher scores mean more recent activity detected.

The ingestion pipeline runs every 12 hours (02:00 and 14:00 UTC). The 30-day scoring window changes slowly, so 12-hour intervals capture daily news cycles without burning excess compute. Scores combine fresh incidents with decayed historical pressure and asymmetric continuity rules.

Structured APIs (USGS, GDACS, ReliefWeb, WHO, GDELT) plus curated RSS feeds from major news outlets, humanitarian orgs, and conflict monitors.

The system detects signals, not ground truth. High-profile conflicts are well-tracked; localized civil unrest may lag. Accuracy varies by country based on source coverage.

Possible reasons:

• Event not yet in our feeds
• Duplicate detection merged it with similar incident
• Keywords triggered demotion (sports, anniversaries)

Note: Score increases appear immediately, but decreases are gradual (max −8/day) to prevent sudden drops during de-escalation.

The map and docs are free to view. For commercial API access or white-label licensing, contact shaw@worldchaosmap.app.

Not yet. A read-only API for scores and incidents is planned. Check back on the Changelog for updates.

Every news item passes through a local NLP engine that extracts event type, severity, content kind (hard incident vs analysis vs context), country roles (location vs actor vs subject), and casualties. This runs entirely in-process with zero external API cost. The classifier uses keyword patterns with contextual analysis — procurement, diplomacy, and policy news is demoted even when military keywords appear.

The system detects 12 Unicode script families (Arabic, Cyrillic, CJK, Devanagari, etc.) and sub-classifies Latin-script text into English, Spanish, French, German, Portuguese, or Turkish. Non-English content is flagged for language-aware processing, and diacritic normalization enables cross-language deduplication.

Each source accumulates a health score across ingestion runs, computed from success rate, incident yield, dedup survival (original vs rehashed content), and response latency. New sources start neutral (0.7) and must prove themselves over 5+ observations. Low-credibility sources get their severity dampened automatically.