Skip to main content

Ontology

LLMO is not page-centric. It is entity-claim-source centric. The following objects form the ontological foundation of the system.

Entity

A company, person, product, service, institution, domain, or concept that must be represented accurately across machine systems. An entity is the primary subject of optimization, operations, and orchestration. Every claim, source, and representation in the system is anchored to one or more entities.

Claim

A specific assertion about an entity. Examples:
  • “Company X is SOC 2 compliant.”
  • “Product Y integrates with Salesforce.”
  • “This API supports structured outputs.”
  • “Jane Smith is the CTO of Acme Corp.”
Claims have states:
StateDescription
TrueThe claim is currently accurate and verifiable
FalseThe claim is factually incorrect
OutdatedThe claim was true at a prior point but is no longer current
AmbiguousThe claim is subject to interpretation or incomplete context
SupersededA newer authoritative claim has replaced this one
UnverifiableThe claim cannot be confirmed from available sources
Claims are the atomic unit of truth in the LLMO system. Optimization ensures they are visible. Operations ensures they are correct. Orchestration ensures they are actionable.

Source

Where a claim comes from. Sources have attributes:
AttributeDescription
OriginThe system, document, or person that produced the claim
Trust levelThe assessed reliability of the origin
TimestampWhen the source produced or last confirmed the claim
OwnershipWho controls or is accountable for the source
Proximity to truthHow close the source is to the authoritative origin of the fact
Conflict likelihoodThe probability that this source contradicts other sources
RetrievabilityWhether the source can be accessed and verified by machines

Provenance

The trace from source to claim to representation. Provenance answers: Where did this claim come from? Through what chain did it reach the model? Was anything lost, distorted, or compressed in transit? Provenance is essential for audit, trust, and debugging. Without it, a representation is an assertion without evidence.

Freshness

Whether a claim is still current. Freshness is not binary. A claim may be:
  • Current and verified within a defined window
  • Stale but likely still accurate
  • Stale and likely outdated
  • Explicitly expired or superseded
Freshness determines whether a model should rely on a claim, flag it for review, or discard it.

Supersession

Whether a newer authoritative claim invalidates or replaces an older one. Supersession is directional. A superseding claim must come from a source of equal or greater authority than the claim it replaces. Supersession chains create a versioned history of truth about an entity.

Representation

How the entity is ultimately compressed and surfaced inside a model-mediated answer or workflow. Representation is the output. Everything else in the ontology exists to govern the quality, accuracy, and trustworthiness of the representation. A faithful representation is one where the claims are current, the sources are authoritative, the provenance is traceable, and the freshness is within bounds.