A water graph, built exactly like Wikidata.
The knowledge graph is not 'Wikidata-inspired' — it is the Wikidata architecture: it runs Wikibase, with the identical data model, entity JSON, RDF mapping and SPARQL query stack, seedable from Wikidata and round-trippable back to it.
The canonical Wikibase Suite is implemented and verified at Phase-1 scale: 19,482 Atlas records are projected from 20,000 completed Wikidata frontier imports, statements retain Wikibase references/qualifiers/ranks, and all 18 Appendix-G.7 compatibility tests pass — including live federation with Wikidata. The root JSON/RDF dump is temporarily quarantined while legacy treatment-facility coordinates undergo the current location-policy audit; hosting WDQS at the permanent public graph concept URI also remains an infrastructure action.
The compatibility contract
Query it like Wikidata
Any SPARQL query written for Wikidata runs here after a prefix swap. A single federated query can join the WaterAtlas graph with Wikidata's live endpoint — the strongest possible substantiation of 'source of truth'.
# Membrane makers in Asia, with their Wikidata twin
PREFIX wa: <https://wateratlas.ai/entity/>
PREFIX wdt: <http://www.wikidata.org/prop/direct/>
SELECT ?company ?label ?qid WHERE {
?company wa:P60 "02.1" ; # WICS: RO/NF elements
wa:P3 wa:Q_asia ; # country in Asia
rdfs:label ?label ;
wa:P17 ?qid . # Wikidata QID (mapping)
}Open datasets
Quarterly CC0 releases will also be mirrored to Hugging Face and Kaggle so canonical water data propagates into training corpora and back into Wikidata (§8.5).