How the atlas is built

Entity model

The atlas treats every concept as a first-class entity with its own URL: wars, battles, factions, and commanders. Cross-cuts — era, theater, year — index the same battles from different angles, so a reader can pivot from 1944 into the Pacific theater into a single commander's career.

Every entity carries a Wikidata QID where one exists. The QID is the join key: it lets us merge data fragments across sources without inventing identifiers of our own and without colliding with similarly-named entities.

Source pipeline

Three upstream sources feed the atlas:

• Wikidata — structured facts (dates, coordinates, belligerent QIDs, hierarchy). Pulled via SPARQL queries against the public endpoint. License: CC0.
• Wikipedia — lead-paragraph extracts and longer prose, fetched via the public REST API. Always attributed to the source article; quoted text retains its CC BY-SA 4.0 license.
• Wikimedia Commons — hero images and battle maps. Each image carries license, author, and source URL pulled from the file's metadata. License-incompatible files are filtered out before ingest.

Ingestion runs as a set of idempotent Django management commands. Each command targets one enrichment slice — discovering a war, attaching a hero image, matching a commander to a QID — so any pass can be re-run safely after fixes or upstream changes.

Curation

Automation handles the long tail. The originals — the wars at the core of the catalogue — are hand-curated: their summaries, causes, and outcomes are written or edited by a human. Where a longer essay appears for an algorithmically-discovered war, the prompt that generated it is constrained to operate on the Wikidata-derived facts already on the page, with inline entity links back into the atlas, and the result is reviewed before publication.

Where coverage is thin we say so. Wars marked scaffolded have the structural data — date range, belligerents, location — but no curated narrative yet. They appear in the catalogue because suppressing them would be its own kind of editorial misrepresentation.

Limits

• The atlas is an index, not the primary literature. Where a claim matters, follow the citation through to the source.
• Coverage skews to conflicts with strong Wikipedia / Wikidata presence. Smaller, regional, or pre-modern wars are systematically under-represented relative to their historical significance.
• Casualty numbers, force strengths, and territorial extents are estimates with wide uncertainty bands. We surface a range when sources disagree rather than picking a number.

Reuse the data

The structured catalogue behind the atlas — battles with dates, coordinates, and source-cited force-strength and casualty ranges — is published as CSV and JSON downloads under CC BY 4.0 on the open dataset page. Structured facts only; Wikipedia prose stays under its own CC BY-SA license and is excluded from the export.

Corrections

The atlas improves when readers point at things that are wrong. Use the contact form for factual corrections, missing battles, or methodology questions.