Methodology

This page documents how OutbreakWatch collects, processes, and presents data. We publish this to make our work auditable and to support the trust signals required for any health-information service in 2026.

Source hierarchy

We rank sources by authority and prefer the highest tier available for each claim:

1. Tier 1 — Official health authorities: WHO Disease Outbreak News (DON), ECDC threat assessments, CDC, Africa CDC, PAHO, national ministries (e.g. Ministero della Salute).
2. Tier 2 — Peer-reviewed research: PubMed-indexed journals, preprints with caveats noted.
3. Tier 3 — Major journalism: Reuters, AP, BBC, NPR, CNN, Al Jazeera, Time, Le Monde, Corriere della Sera, ANSA, RAI News, ARD, Le Figaro.
4. Tier 4 — Trade journalism and aggregators: STAT News, ProMED, Healio, Quotidiano Sanità.

We exclude blog/personal opinions, anonymous Telegram/4chan claims, and unverified social media reports.

News ingestion

We ingest RSS feeds from the sources above every 4 hours via a Cloudflare Worker. For each new item we:

1. De-duplicate by canonical URL and headline similarity (cosine on embeddings).
2. Tag with one event type: confirmed_case, death, tracing, response, media, science.
3. Extract location with a Gemini 2.0 Flash model prompted with strict schema; fallback to Nominatim (OpenStreetMap) for geocoding.
4. Validate location precision: exact > city > region > country.
5. Reject items with no extractable location or with location confidence below threshold.

Prediction-market ingestion

We poll public APIs of Polymarket and Kalshi every hour. We store the YES probability and total dollar volume snapshots. We display the live values plus a 7-day history curve. We never display order book details or facilitate trades. Markets are linked back to their original platform for users who wish to participate.

Per-country aggregates

Country pages aggregate confirmed cases, suspected cases, and deaths from primary sources only. Risk level (none / very low / low / moderate / high) reflects the published assessment of the most relevant authority (ECDC for Europe, PAHO for the Americas, Africa CDC for Africa, etc.). We never assign a risk level by editorial fiat.

Editorial review

AI-generated content (summaries, FAQ answers, comparisons) is reviewed against the cited primary sources by a human review pass before publication. Each piece carries a last updated timestamp. Corrections are dated and visible at the bottom of affected pages.

What we do NOT do

• We do not invent numbers. Every figure is linkable to a source.
• We do not generate symptom self-checkers, triage tools, or individual medical advice.
• We do not aggregate user-reported symptoms or run citizen-science data collection.
• We do not push notifications or alerts unless explicitly opted into by the user.

Updates and corrections

Significant errors are corrected within 24 hours of being identified. Corrections are marked at the bottom of affected pages. To report an issue, email corrections@outbreakwatch.net.