The Negative Event Density Index (NEDI) quantifies the "reporting density" of historical crises by counting distinct negative incidents documented per year in Wikipedia's chronological archives.

• Data source: raw wikitext for each year page, fetched via the MediaWiki revisions API in real time.
• Event isolation: only the Events section of each year page is parsed — Births, Deaths, Awards and reference sections are excluded at the source.
• Keyword filtering: entries are matched against crisis identifiers (war, attack, massacre, pandemic, coup, famine, earthquake, and ~40 more) using whole-word matching to avoid substring false positives.
• Strictly negative: response and relief actions (recovery funds, vaccines, aid, peace talks) are excluded even when tied to a crisis, and entries that inherit negativity only from a parent bullet (e.g. an election listed under "COVID-19 pandemic") are dropped when procedural.
• Event classes: each incident is tagged human (violence, political, economic, technological) or non-human (geophysical, weather, disease) from its matched vocabulary; incidents matching both appear under both filters.
• Death toll (estimated): tolls are parsed from event text ("killing at least 944", "death toll exceeds 500,000"). Incident figures are summed per year; cumulative milestones take the year's maximum to avoid double-counting. Charted on a log scale. Coverage is partial — treat as a lower bound.
• Severity colouring: Relative mode colours bars by standard deviations from the range mean, highlighting true outlier years; Heatmap mode uses a continuous gradient over absolute counts, so mid-range years read as warm rather than neutral.
• Trajectory: the verdict compares mean incident volume in the second half of the loaded range against the first half, alongside a least-squares linear trend (R² shown). It inherits the active event-class filter.
• Local caching: parsed years are cached in the browser with tiered expiry — 24 hours for the current era, 7 days for the last decade, 30 days for historical years — so article updates are picked up on the next expiry. Shift+click Run Analysis to bypass the cache entirely.
• Density vs. sentiment: NEDI counts the volume of distinct reported incidents, not their subjective severity.