Data & measurement
How the automatic analysis is calculated
The public inputs, sample limits, exclusions, update behavior, and interpretation rules behind the analysis page.
A transparent report names its denominator, its time boundary, and its blind spots. A zero means something only when the source was successfully read.
Read only existing public data
Current signals come from WOUT’s public intelligence-event view. The request takes up to 200 of the most recently reported unexpired records, plus one extra row to detect a limited sample. The response is validated and projected to the fields shown on this site.
Recent news comes from the existing public news interface, limited to up to 50 returned items. Published items within the preceding 48 hours are analyzed; returned retractions are kept distinguishable and excluded from the published-news count. The 24-hour security figure comes from its existing public aggregate.
Keep each metric’s denominator visible
The dashboard reports category counts, evidence states, place labels, geographic precision, and the median age of the latest report attached to each signal. Place, category, evidence-state, and expiration filters apply to the charts and record lists; headline metrics retain the full returned-sample denominator. A resolved record can remain an unexpired published record; it must not be called an ongoing incident.
Hourly bars group the latest report attached to each currently returned signal, not every report ever received. They show six clock hours in Haiti time, including the partial current hour. Expiration bars group remaining display time into under one hour, one to three hours, three to six hours, and six hours or more. Place counts use the supplied place label, falling back to the area identifier or Location unresolved. Sample limits and invalid records are disclosed. News publication names are not automatically independent witnesses. Current signals and the historical security metric use different definitions and should not be compared as interchangeable totals.
Refresh without manufacturing history
The page requests a fresh calculation every two minutes while it is open. Server-side source reads may be cached for up to two minutes. Signals that expire are removed from the displayed calculation, even if the underlying source response was cached.
All-time analysis uses a separate aggregate of the retained publication ledger when its database interface is connected. It includes eligible expired events, discloses the recording start and original backfill limits, and never substitutes the current feed for historical data. See Logistics, history, and saved briefs for definitions.
An unavailable source is not zero
Signals, news, and the security aggregate have separate health states. An unsuccessful read shows Unavailable for that input rather than a zero. A successful empty response is described as an empty WOUT sample, not as a safe or inactive area.
When the whole refresh fails, the page identifies any retained snapshot as the last successful calculation. Source links are limited to normal web URLs, and all titles are displayed as text.
What should be measured next
Useful next measurements include publication latency, correction and dispute resolution times, the rate of ambiguous cluster matches, source availability, and how often cited answers are actually supported by their cited records.
These are evaluation priorities, not benchmark results. This documentation does not invent uptime, audience size, coverage percentages, accuracy scores, or measured user outcomes.
Based on the WOUT implementation reviewed October 11, 2026. Rules and product availability can change. Examples are explanatory; current records appear in the public analysis.
Inspect the latest public calculation