Skip to main content

Trust Graph

The trust graph is the persistent structure connecting subjects, producers, contexts, signals, and relationships in PTI.

Graph model

Node types

NodeKey attributes
Subjectpti_id, identity_type, verification_level
Producerproducer_id, entitled contexts
Consumerconsumer_id, lookup tiers
Contextcontext_id, context_tier
Signalsignal_id, polarity, weight, effective_at
Evidenceevidence_id, evidence_type, hash

Edge types

EdgeSemantics
EMITTEDProducer → Signal
AFFECTSSignal → Subject
SCOPED_TOSignal → Context
SUPPORTED_BYSignal → Evidence
ENDORSESSubject → Subject
EMPLOYSOrganization → Individual
VERIFIED_BYAssertion → Producer

Edges SHOULD carry valid_from and optional valid_to for temporal queries.

Temporal history

The graph is inherently temporal. Intelligence engines query signals as-of a point in time for audit replay:

signals(S, C, t) = { σ | σ.AFFECTS = S ∧ σ.SCOPED_TO = C ∧ σ.effective_at ≤ t ∧ ¬expired(σ, t) }

Historical replay supports dispute investigation and regulatory examination.

Trust relationships

A trust relationship is a durable edge between subjects or between subject and institution:

  • Employment relationship (organization employs individual)
  • Landlord-tenant relationship
  • Counterparty partnership
  • Community endorsement

Relationships may unlock relationship-scoped signals (e.g., employer verification) without exposing unrelated contexts.

Graph isolation

Context isolation is enforced by constraining traversals:

  • Lending lookups traverse lending-scoped signals unless lens derivation rules expand scope.
  • Cross-context inference MUST be explicit in lens derivation_rules.

Query patterns

PatternUse case
NeighborhoodAll signals for subject in context
Provenance walkSignal → event → producer
Path trustEndorsement chain depth and strength
Coverage analysisDetect thin subgraphs

Federation

Federated deployments replicate subject nodes and foreign assertion edges. Conflict resolution prefers higher-confidence registry merges and signed assertion precedence.

Storage considerations

Implementations typically combine:

  • OLTP store — identity and entitlement records
  • Event log — immutable event stream
  • Graph index — traversal-optimized signal adjacency
  • Derived cache — materialized context scores