Patient Context Graph

Everything the Clinia Context Engine knows about a patient is held in one graph.

This graph is used to render the agent-optimized VFS. It's also accessible for auditing and debugging. The following node and edge types can be fetched from the patient context graph.

Everything below the dashed rule is the graph. The faded boxes above it are where records and conversations come from; they are not node types and cannot be fetched.

The extracted prefix marks the parsed, but un-processed version of the inputs. These data points allow you to understand what was kept and what was dropped by the parser.

Fact

One patient axis — a condition, a medication, a lab, an allergy. Structured data is parsed as extracted facts; resolution decides which of those are the same thing and projects them into the one fact that the VFS and condition stories read.

Narrative

Free-form text — a FHIR DocumentReference, a CDA section, a clinical note — is parsed as an extracted narrative and chunked into narrative passages.

Resolution projects each resolved cluster of passages into one narrative, a single titled account of what that cluster says.

Clinical event

A point-in-time record of something that happened: an encounter, an observation such as a lab result, a procedure, or an administered dose.

Learned memory

A statement kept from a chat session — a symptom, an adherence report, a concern, or a preference.

Resolution decision

One recorded verdict per compared pair: merge or reject, the rule that fired, and the evidence each layer of the resolution cascade produced.

Resolves to

Points an extracted entity to its canonical representation. Walking these edges backwards from a fact or a narrative gives you every source record behind it.

Relationship

A typed, directed clinical link: (full list).

Chunked into

Links an extracted narrative to the passages its text was split into.