Contested evidence prototype
A working research-infrastructure prototype that binds politically sensitive claims to sources, locators, viewpoints, legal and policy records, and visible human-audit states.
Open the working prototypeResearch problem
Evidence in politically sensitive conflicts does not circulate in a neutral space. A factual statement, legal position, state narrative, policy response, or historical interpretation can describe the same event while depending on different institutions and standards of proof. Conventional narrative pages rarely expose that structure.
This working prototype tests whether those dependencies can be made inspectable. Its current pilot concerns the Lachin corridor blockade, the September 2023 displacement from Nagorno-Karabakh, and the EU and international response. The case is deliberately bounded; the system is not presented as a definitive history or a verified scholarly database.
Implemented architecture
The relational model connects events, places, people, parties, policies, legal rulings, heritage sites, sources, disputes, and viewpoints. Claim-level citation anchors connect prose to source records and precise locators. Each claim can carry a role, status, source set, competing viewpoints, an editorial audit, and a human-edited revision.
The SvelteKit interface exposes those relationships through citation popovers, bibliographies, graph and map views, revision history, completeness metrics, short links, and research-friendly CSV exports. PostgreSQL stores the records and audit trail; server-side loaders resolve citations and apply verified revisions without silently replacing the original record.
Verification model
AI-assisted drafting is treated as provisional material, not authority. A claim earns a verified state only after a person checks the source, locator, wording, and context. Other states—including corrected, disputed, rejected, unresolved, and unaudited—remain visible so that disagreement and correction are part of the method rather than hidden editorial residue.
The next research phase narrows this into reviewable claim packets: claim text, sources, locators, current status, possible viewpoints, reviewer notes, and explicit confirm, correct, dispute, reject, or defer decisions.
Why it matters here
The work extends beyond cartography. It is an experiment in research infrastructure for source-bound claims, human-audited provenance, uncertainty, and correction. The Armenia–Azerbaijan case is the pilot through which that method is being tested, not the final intellectual boundary of the system.