The QIS knowledge graph.

Structured documents are published at stable URLs, served with permissive CORS headers and long cache lifetimes. They are versioned, they will not move, and they are meant to be consumed by other people's software:

• /api/graph.json — the full node and edge set, with a fixed layout, the node-type and relationship-verb ontology, and each node's own context-bundle URL.
• /api/graph/context/<id>.json — one file per node: its bounded two-hop neighborhood, with term definitions inlined, so a retrieval system fetches one small document instead of the whole graph to reason about one thing. id is any node id from graph.json, with : and / replaced by -.
• /api/pages.json — every canonical URL across the three domains, with titles, descriptions, layers, and topics.
• /api/glossary.json — the canonical QIS definitions as structured terms.

The same graph is also published as real W3C RDF, not a JSON shape a consumer has to be taught: /api/graph.ttl and /api/graph.jsonld are every node and edge as RDF individuals and typed triples, each node's own real URL as its subject; /api/ontology.ttl is the class and property hierarchy behind them, including verified alignments to PROV-O, SKOS, and one FIBO class; /api/shapes.ttl is SHACL — the structural rules this build already enforces, restated so an external validator can check them independently. All four are re-parsed and SHACL-validated on every build, not merely written and trusted.

All are also described in llms.txt and llms-full.txt for retrieval systems. Every one of these is static, precomputed at build time, and served from the edge — no query surface, no runtime compute, nothing to rate-limit. Attribution to the source URL is the only thing asked in return.