Cross-cutting connections between units, mineral occurrences, sources, nomenclature, rock and chronology terms — Carpathian–Balkan region.
Every node here traces back to a source article (article_id) and, where available, the
exact excerpt it came from — so any connection can be checked against the original text instead of
taken on faith. The raw records stay heterogeneous on purpose (a geological unit and a mineral occurrence
don't share one schema) instead of being forced into a single table. This graph is what makes that messy,
partly-incomplete, multi-source corpus easier to query — by a person clicking around, or by an LLM /
machine-learning pipeline reading the same JSON later. Deterministic tables are still the backbone
(Lexicon, Stratigraphy) and always will be — this
view sits on top of them, it doesn't replace them.
Not included here: spatial contacts and geometry-based relations (overlies/underlies/crosses, map polygons, faults) — that graph is computed from real coordinates and lives in Stratigraphy; open a unit there for its measured neighbours. Rock-class and geological-age links below are a small best-effort keyword match on the extracted text, not the full Rocks / Chronology reference trees — a missing link is not evidence the connection doesn't exist.
Edge weight (shown as ×N in the relations table) counts how many times the same connection is backed by evidence — e.g. a rank shared with N other units, or a commodity reported in N occurrences — not a measured quantity.