Sources and methodology

Source selection

Worldwide Outbreak has the owner’s written permission to use WHO RSS feeds. The reader reproduces text delivered in those feeds with WHO attribution. CDC agency-authored feed material and ECDC material are used under their applicable terms. Source credits and licences remain available in each story’s publication details.

What automation does

The source poller retrieves bounded RSS responses, validates their structure, normalises source dates to UTC and checks for duplicates and content changes. It uses conditional requests where supported and backs off after failures. It keeps the last verified snapshot available with clear freshness information.

What automation does not do

A successful poll does not create a new article when nothing meaningful changed. We do not infer case counts, medical advice or personal risk from headlines. Topic labels are generated from source headline terms and can be imperfect. Travel notices are not a complete global outbreak census.

Dates and corrections

Observation, publication, source update, retrieval and local editorial change are separate concepts. Legacy CMS dates are labelled when their meaning is uncertain. Corrections should retain the reason, evidence and date of the change. Sources remain directly accessible from the relevant page.

The data explorer

The COVID-19 explorer uses WHO’s weekly machine-readable dataset, published under CC BY 4.0. Values are aggregated from country, area and international-conveyance rows for the same reporting week. The reporting date and retrieval date are different. Country names containing damaged accent characters in the source are normalised for display; numerical observations are unchanged.

The explorer also separates Ebola, dengue and mpox datasets. Each series retains its source, reporting period, geographic scope and measure; a cumulative total is not presented as a current weekly case count. Country-level notices and chart coverage do not establish an individual's infection risk. Missing data is not treated as proof of zero cases.