Core Concepts

Trust Layers

Not all evidence is created equal. A raw lab result from Quest Diagnostics is not the same as an AI-predicted treatment response. OpenBio makes this distinction explicit on every assertion with a six-level trust hierarchy.

source_factSource Fact

Raw, unmodified data directly from the source system

normalized_factNormalized

Mapped to standard terminology (LOINC, SNOMED, ICD)

extracted_annotationExtracted

Derived from unstructured documents or images via NLP/AI

derived_featureDerived

Computed from other evidence assertions

model_outputModel Output

AI/ML model predictions with confidence scores

hypothesisHypothesis

Proposed clinical or research interpretations

Filtering by trust layer

Always specify the trust layers you want. High-trust queries exclude model predictions; exploratory queries can include them.

typescript

// High-confidence clinical data only
const clinical = await client.evidence.bySubject({
  subjectId: 'james-58',
  trustLayer: ['source_fact', 'normalized_fact'],
});

// Include AI predictions for exploratory analysis
const exploratory = await client.evidence.bySubject({
  subjectId: 'james-58',
  trustLayer: ['source_fact', 'normalized_fact', 'model_output'],
});

// The trust layer is always present on the response
clinical[0].trustLayer; // → 'source_fact'
clinical[0].confidence; // → 0.99