Methodology

How people are selected

The primary source is Wikipedia’s “Lists of deaths by year” and the linked yearly/monthly pages. People included in those lists are candidates for the archive. Additional reliable sources may be added in the future.

How Wikipedia data is imported

Data is imported through permitted programmatic access to Wikipedia/Wikimedia APIs rather than uncontrolled HTML scraping. For each record we store the Wikipedia page URL, page title, source URL, import date, last-verified date, source license and, where available, the Wikipedia revision ID and revision timestamp.

Duplicate handling

The same person may appear in multiple yearly or monthly lists. We detect duplicates using Wikipedia URL, Wikidata ID, canonical name, date of birth and date of death as matching signals. Two people are never merged solely because they share a name.

Date normalization

Dates are stored internally in ISO format (YYYY-MM-DD) and displayed in human-readable English. Partial dates are supported where the exact day is unavailable; missing dates are never fabricated.

Occupation & country categorization

Occupations and countries are categorized using deterministic rules and predefined mappings, not AI classification. Each person can carry multiple occupations.

Image selection & license verification

Preferred image sources are Wikimedia Commons and Wikipedia/Wikimedia metadata. For each image we store the image URL, source URL, creator, license, license URL, attribution text and verification status. If a suitable license cannot be verified, the image is not displayed and a neutral placeholder is shown instead. We do not use commercial stock photos without an appropriate license, and we do not generate images using AI.

Statistics

All dashboard statistics are calculated directly from the database using deterministic queries (count, average, median, grouping). Where the underlying data is incomplete, those records are excluded from the relevant calculation rather than estimated.

News verification

“Who Died Today?” is a separate layer from the historical archive. Candidate articles are collected from permitted news feeds and APIs using deterministic keyword and named-entity matching. Corroboration from more than one reputable source is required where possible; single-source entries are clearly labelled and never presented as confirmed fact without verification.

Corrections

Readers can report errors from any person page. Submitted corrections are reviewed by an administrator before any change is published; user-submitted information is never exposed publicly without moderation.