The atlas is built around one fact (the September 2023 expulsion of the entire Armenian population of Nagorno-Karabakh) and pulls the dossier outward from there: the events, figures, places, parties, policies and legal rulings that produced it, plus the symmetric record of Azerbaijani grievance on which it is partly built. Nothing is flattened. Where readings of an event diverge, every reading is presented in its own voice and its supporting sources are named.

The atlas is also relational: every entity links to every other, so a reader who starts at the 1992 Khojaly massacre can step in either direction (back to the Sumgait pogroms that preceded it, forward to its instrumentalisation in 2020s Azerbaijani state rhetoric) without leaving the structure. The graph view exposes that web directly.

Author

David Wicker — wicker.life

Independent researcher; not affiliated with any state, party, or institution involved in the dossier. Errors and omissions are mine. Corrections are welcome: david@wicker.life.

Online discussion of the Armenia-Azerbaijan dossier moves fast and rarely cites. A claim is made, a counter-claim follows, both sides feel they have history on their side, and the readers most affected by the outcome (Caucasian, Armenian-diaspora, Azerbaijani-diaspora, and policy audiences in capitals making energy and security choices about the region) often don't have a single place to check what is documented, what is contested, what is sourced, and by whom.

The atlas is that single place, in working draft form. It tries to be useful in three different reading modes: as a topic encyclopedia (start anywhere, follow the links), as a debate companion (the rebuttals page lists common arguments and their sourced responses), and as a research scaffold (every claim has a source row that lists author, title, year, publisher, and where applicable the URL).

Three commitments shape the atlas's editorial choices.

Sourced over asserted. Every claim is referenced to an underlying source: books, court rulings, UN resolutions, NGO investigations (HRW, Amnesty, Memorial, ICRC, Caucasus Heritage Watch), primary documents and academic monographs (Hovannisian, de Waal, Suny, Bournoutian, Kévorkian, Akçam, Swietochowski, Altstadt). Where a claim is editorial synthesis rather than directly cited, it is marked with the editorial chip; cited opinion is marked sourced opinion; genuinely disputed claims are marked contested.

Symmetric over partisan. Maximalist Armenian and maximalist Azerbaijani framings are addressed in their own voices on the disputes and rebuttals pages, with a separate critique block under each that names omitted facts, logical fallacies, or rhetorical moves. The atlas is not symmetric in its conclusions, because the historical record is not symmetric, but it applies the same evidentiary standard to a maximalist Armenian claim and to a maximalist Azerbaijani claim. War crimes by either side are named as such.

Ranges over false precision. Where casualty counts or displacement figures are contested, ranges are given (e.g. "100,000–120,000 displaced from Karabakh in 2023") rather than picking a single number. Where a single number is well-established (e.g. UNHCR's 100,617 registered displaced), it is used and cited.

The text-search engine and the relational graph are both built on top of this structure so that following a link, finding a source, and seeing the relations between entities always lead back to the same evidentiary base.

Palestine Atlas

A parallel atlas of the Israel–Palestine record built on the same editorial principles lives at palestine.wicker.life. Different conflict, same commitment to sourcing, symmetry, and contestability of claims. The two atlases share infrastructure but no data; they are separate dossiers.

The atlas is a working draft. The structural frame (entities, relations, sources, perspectives) is stable; the content density varies by topic — some eras are well-fleshed-out, some are sketches that will be filled in over time. The completeness scoreboard shows where the gaps are, by entity type. If you find a missing claim, an unsourced assertion, or a factual error, please send the correction.

Built with SvelteKit and Postgres (Neon). Maps are Leaflet over CartoDB Positron tiles. The relational graph is rendered with d3-force. Search is Postgres FTS with pg_trgm fuzzy fallback. All atlas content is in a single relational store; the rendered pages are streamed and edge-cached at the CDN.