Global Health Observatory

Coverage and source families

The initial catalogue contains 199 endpoints across 29 agency/publisher families: 159 RSS/Atom channels, 39 structured data endpoints, and one adapter reusing Watchtower’s reviewed health-publisher snapshot. The source catalogue reports actual collection outcomes, publication or observation periods, polling cadence and bounded samples. Channel counts are not independent source counts: CIDRAP’s 106 disease/topic channels remain one publisher family.

Sources include WHO Disease Outbreak News and current news, WHO FluNet and GHO, World Bank indicators, CDC surveillance/publications, ECDC, PAHO, WHO Africa, UKHSA/MHRA, EMA, FDA, NICD, Taiwan CDC, RIVM, SSI, ANSES, Fiocruz, WOAH, specialist reporting, scientific publishers, Europe PMC and ClinicalTrials.gov. This is a curated, source-bounded collection, not an exhaustive record of global disease events.

The research ledger checked 280 candidate URLs. Unavailable, superseded, empty, challenge-blocked and access-gated endpoints are visible as research gaps. GISAID, detailed WAHIS/EMPRES data and commercial providers require their own access/schema/redistribution review and do not contribute licensed observations to this shared dataset.

Different evidence types

• Official notices retain agency provenance and publication dates. A notice or headline does not independently establish every reported claim.
• Reporting and research retain source links and bounded excerpts. Journal metadata does not establish peer review, reproducibility or a successful clinical result.
• Clinical studies expose registry metadata, phase/status and reported study countries, not patient records, successful outcomes, approvals or inferred facility coordinates.
• Regulatory recalls retain classification and status, including terminated recalls. Report dates are not automatically incident dates; distribution text is not converted into case locations.
• Measurements retain the original reporting period, unit, surveillance stream, jurisdiction and population stratum. Annual baselines are not real-time hospital capacity.

Surveillance measurements

• WHO FluNet supplies recent weekly virological records. Positive specimens, specimens processed and surveillance origins remain distinct; these are not population case totals.
• CDC wastewater WVAL and H5/mpox/measles detection records are limited to State_Territory and CDC_Verily provider designations. WastewaterSCAN rows, which carry separate non-commercial conditions, are excluded at query and parser levels. Environmental detection does not identify human infections. No treatment or laboratory-method instructions are collected for display.
• CDC NHSN retains admissions, rates per 100,000 and hospital reporting coverage separately. RESP-NET retains network, age, sex, race, jurisdiction, weekly/cumulative and observed/adjusted strata. State and national rows are never summed.
• World Bank annual indicators cover workforce/capacity, expenditure, immunization, mortality, HIV, malaria, TB, water, sanitation, stunting and PM2.5 exposure. Regional aggregate rows and null/suppressed values are excluded from country measurements, rather than becoming zero.
• WHO workforce values retain per-10,000 units. IHR SPAR fields are country self-assessment scores, not independently established readiness. Older AMR indicators retain their source-reported scale and vintage; collection freshness does not update a survey year.

Charts and exports preserve strata. A country with multiple entities/sites is not assigned an averaged national value. Country shading uses the latest single value for the selected series/entity context; missing or multiple values remain unshaded. Cross-country latest periods can differ.

Geographic meaning

Country/territory reference points identify reported country codes, study-country metadata or explicit publication-title country mentions. Overlapping names and ambiguous names are handled conservatively; unsupported locations remain unmapped. Publisher headquarters are not substituted for event locations. Points and country fills are not outbreak footprints, individual case locations, infected populations or transmission estimates. Publication activity charts count linked records, not infections.

The separate downloadable country reference contains information from mledoze/countries, made available under the Open Database License 1.0 (license copy). The reference derivation and source-content hash are included with it. Natural Earth 1:110m boundaries are public domain. Reference geography is not a statement about disputed territorial status. Health source records retain their respective provider terms.

Collection, dates and recovery

The background job checks due sources every 30 minutes; individual sources normally poll every 2, 6 or 24 hours. Host-aware concurrency, fixed HTTPS hosts, pinned public DNS, response/time budgets, conditional validation and provider backoff bound collection. Reader queries use the stored snapshot and never trigger arbitrary provider collection.

Publication time, observation period, source validation and collector time remain distinct. Missing/invalid/future publication dates are not replaced with the retrieval clock. HTTP304 validates unchanged content without making its publication newer. Rarely updated feeds can be available while containing old publications.

Publications are bounded to 60 rows per RSS channel; WHO notices/current news to 100; clinical-registry pages to 300 studies; Europe PMC to 300 metadata records. WHO FluNet uses roughly 12 weeks, CDC wastewater roughly five weeks, hospital/RESP-NET roughly 13 weeks, annual series normally six prior years plus the current year, and older AMR series a longer historical window. Provider pagination has explicit page/row caps; capped sources are marked. Canonical links are deduplicated with carrying-source provenance retained. These limits are a working collection, not a complete historical archive.

Failed sources can retain original-dated last-good records for seven days for publications or 30 days for measurements, marked stale/unavailable. Source-state publication is atomic. Fourteen daily compressed snapshots are retained operationally. Collection snapshots older than two hours are labeled stale in the reader.

The primary reader remains usable during account checks. Prior selections and pending responses are owner-fenced, and classic provider controls are withheld until verification completes. Failed same-selection refreshes retain accepted data; other failures show retry controls. Live observatory requests and results are withheld during replay, with classic captured analysis accessible separately. Graphics failures retain the data explorer and source catalogue.

Public data interface and exports

GET /agent/api/health-world supports reviewed source, country, topic, pathogen, evidence kind, literal text, publication window, series/entity and bounded pagination filters. Responses include source health, provenance, map references and reporting periods. Unsupported parameters, private selectors, caller URLs and forced collection are rejected. Exports contain the selected page and its metadata, not an unbounded provider dump.