TrustRank Assessment Methodology
TrustRank is an independent, multi-dimensional assurance system designed to evaluate AI agents under defined configurations, permissions, and empirical evidence.
Multi-Dimensional Scoring
Evaluates technical architecture across deterministic dimensions: system security, runtime containment, behavioral idempotency, intent binding, memory persistence, and financial authority limits.
Evidence Tiers (E0–E3)
Every scoring decision requires documented evidence. Tiers range from unverified provider claims (E0) to public documentation (E1), empirical test telemetry (E2), and cryptographic independent verification (E3).
Control Findings (CF1–CF4)
Identifies vulnerabilities and policy deficiencies. Critical findings (CF1) trigger automated score capping or immediate revocation, preventing high ratings when fundamental safeguards are missing.
Lifecycle & Provenance
Assessed states are time-bounded. Records progress through valid, expired, suspended, and revoked lifecycles. Artifacts are bound by SHA-256 fingerprints to ensure integrity.
Semantic Precision: Status Distinctions
Invariants governing status presentation and claim boundaries.
| Concept | Definition | Boundary |
|---|---|---|
| CLAIMED | Self-asserted statement submitted by the provider. | Unverified. Zero independent assurance. |
| VERIFIED | Provider identity or technical artifact cryptographically confirmed. | Confirms authenticity, not behaviour. |
| ASSESSED | Evaluated against the TrustRank multidimensional framework. | Yields a bounded Trust Passport record. |
| CERTIFIED | Prohibited Terminology | TrustRank does not certify AI agents. |