QRI is designed to assemble the permitted evidence around an investigation from the systems where it already lives, preserve exact source and receipt lineage, and expose that context to people, software, and agents through the same controlled application boundary.
A closed quality-specific model over exact source cases, reports, facts, evidence, RCA conclusions, document revisions, suppliers, actions, and effectiveness checks. It is not a generic ontology product or a second source of truth.
Durable synchronization and writeback retain source ownership, mapping versions, capture receipts, retries, explicit unknown outcomes, and reconciliation. Connectors—not MCP—remain the integration plane.
Ordinary applications integrate through versioned application contracts. Public APIs expose typed product capabilities without giving external systems direct database or authority access.
MCP is intended for permission-filtered reads and immutable proposals after a customer names a real agent workflow. It never replaces connectors, queries PostgreSQL directly, or exposes model-controlled approval, closure, waiver, effectiveness acceptance, or external dispatch.
The first context is intentionally narrow. It references existing authoritative subjects and exact versions instead of copying business records into a new graph database.
A relationship must retain exact endpoint identity, source event or receipt, mapping version, visibility scope, assertion class, and human disposition. AI proposals cannot silently become authoritative relations.
Connectors, model providers, retrieval services, document processors, SDKs, and MCP processes receive only scoped inputs and capabilities. They do not receive authoritative database credentials and cannot approve, sign, close, waive, release, accept effectiveness, or dispatch external actions.