Source hierarchy

Frontlines prioritizes official government, UN, and institutional sources ahead of multi-source discovery feeds. Social scraping, noisy unofficial channels, and opaque providers are deliberately excluded from the MVP.

SourceCoverageInterval
GDELTDiscovery feed for global conflict, protest, disaster, and infrastructure reporting.15m
GDACSUN/EU disaster awareness and coordination alerts for major hazards.15m
USGSProgrammatic GeoJSON feeds for recent earthquake events.5m
NASA FIRMSNear-real-time fire and hotspot detections from MODIS and VIIRS.30m
NWSOpen weather alerts, forecasts, and observations API for the United States.5m
ReliefWebHumanitarian reports and updates curated from thousands of sources.15m

Signal and impact scoring

FactorWeight
Source quality30
Signal diversity20
Severity20
Recency15
Geo confidence10
Operational relevance5

Signal scores are not predictions. Watchdesk impact scores add proximity, configured radius, watch item priority, and source confidence to decide what deserves operator attention now.

Decision path

  1. Pull and normalize each upstream feed inside its own adapter and record source-run health.
  2. Deduplicate and score stored events before operator-facing filtering and watch-item matching.
  3. Route only deterministic alerts into queue transitions and delivery scheduling.
  4. Use AI only after the stored event set exists, and reject any drafted output that cannot bind claims back to source labels.

Source freshness gates

  • Fresh: Last success is within 2x the expected interval.
  • Delayed: Last success is between 2x and 6x the expected interval.
  • Stale: Last success exceeded 6x the expected interval.
  • Down: Multiple consecutive failures have taken the source offline.
  • Disabled: Manual kill switch disabled this source.

Operating boundaries

  • Alert matching, delivery, and queue transitions remain deterministic. AI-assisted briefs can be enabled for sourced summarization, but alerts never depend on an LLM response.
  • Frontlines provides source-backed operational awareness and does not replace emergency services, government advisories, professional security judgment, or official travel guidance.
  • Every AI-generated brief claim must bind to stored source references. If AI is unavailable or fails guardrails, the product falls back to deterministic templates.
  • Programmatic country and topic pages remain non-indexable until they have enough real coverage and unique analysis to stand on their own.

Methodology FAQ

Does Frontlines use AI in the alert critical path?

No. Alert matching, delivery, and queue transitions remain deterministic. AI only augments sourced summaries and falls back to deterministic templates when unavailable.

What sources does Frontlines prioritize?

Frontlines prioritizes official government, UN, and institutional sources ahead of multi-source discovery feeds, and excludes noisy unofficial channels from the MVP.