Skip to main content

Trust Intelligence Engine

The Trust Intelligence Engine derives context-scoped outcomes, confidence bands, and explainability artifacts from the trust graph.

Engine position​

Inputs​

InputDescription
Active signalsContext-filtered signal subgraph
Evidence weightsVerification level multipliers
Relationship edgesEndorsement and employment links
Policy tablesDecay, caps, lens derivation rules
Historical outcomesOptional calibration datasets

Outputs​

OutputDescription
score_pct0–100 normalized context score
bandthin, fair, good, strong
confidenceModel certainty 0.0–1.0
driversRanked explanatory factors
coverage_gapsExplicit thin-data flags
provenance_summaryProducer diversity and freshness

Scoring pipeline​

Context Score vs Trust Score​

TermScope
Context ScoreSingle context_id outcome
Trust ScoreInformative umbrella term; implementations SHOULD avoid presenting a single cross-context number without explicit aggregation policy

PTI emphasizes context-scoped outcomes to prevent context collapse.

Explainability contract​

Explainability artifacts follow explain_score.v1 structure:

{
  "confidence": {
    "score_pct": 72,
    "band": "good",
    "drivers": [
      {"id": "repayment_on_time", "label": "On-time repayments", "weight": 0.34}
    ]
  },
  "coverage_gaps": ["thin_insurance_history"]
}

Drivers MUST map to auditable signal classes, not opaque latent features, in conformant profiles.

Model governance​

Operators SHOULD maintain:

  • Versioned model cards per context
  • Bias and fairness review cadence
  • Champion/challenger promotion process
  • Rollback path for score distribution drift

Refresh modes​

ModeTrigger
Event-drivenNew materialized signal
ScheduledNightly batch for low-velocity contexts
On-demandLookup requests stale cache

Stale cache SHOULD expose computed_at so consumers judge freshness.

Screening integration​

Screening dimensions (sanctions, PEP, identity registry) MAY attach as adjunct compliance_intelligence blocks without replacing context scores.