Methodology v0

Trust Score today. Entity confidence next.

Trustprint's public v1.1 score is a visible review signal for indexed AI agents. The next layer is evidence-backed cross-registry entity confidence, built conservatively and measured before it changes public scores.

What v1.1 Measures

The current public product is a Trust Score for indexed EVM agent addresses. It is useful for sanity-checking a counterparty before trust, payment, routing, integration, or manual review. It is not an underwriting decision and not a claim that every agent identity is already resolved across all registries.

Component	Weight	Meaning
Activity	0.25	Observed engagement and chain breadth signals.
Longevity	0.20	Time since first indexed registration.
Cross-registry	0.25	Coverage across tracked registries. This is not yet verified entity resolution.
Consistency	0.20	Completeness and stability of visible metadata/signals.
Anomaly-clean	0.10	Early clean-behavior dimension; deeper risk policy remains roadmap.

score = activity*0.25 + longevity*0.20 + cross_registry*0.25 + consistency*0.20 + anomaly_clean*0.10

The Next Layer

Trustprint separates the health of a footprint from the confidence that several footprints belong to the same real operator. That turns chain breadth from a count into an evidence question.

Trust ScoreHow healthy does this observed footprint look?

Identity ConfidenceHow sure are we that multiple footprints represent the same entity?

Evidence Classes

Class	Evidence	Decision Use
Hard	Signed proof-of-control, shared controller wallet, signed metadata pointer, verified LUKSO profile or equivalent anchor.	Can support strong confidence.
Soft	Shared deployer, metadata fingerprint, shared endpoint/domain, correlated timing or behavior.	Candidate signal only.
Weak	Name, logo, or free-text similarity alone.	Never sufficient for merge.

Precision comes before recall. A false merge is worse than a missed merge.

Decision Rules

strongAt least one hard evidence class and no contradiction. May form an anchored entity.

moderateAt least three non-trivial soft signals and no contradiction. Candidate only.

weakOne soft signal or weak-only similarity. Store as hint or ignore for resolution.

conflictHard evidence points to incompatible entities. Split and flag; never guess.

Claim Boundaries

Allowed todayTrustprint indexes and compares agent records across registries. The v1.1 score includes a cross-registry coverage component. Verified entity resolution is the next methodology layer.

Not claimed todayTrustprint does not claim every agent identity is already resolved, that LUKSO is live-indexed, or that the public score is an underwriting/compliance verdict.

Validation Plan

Build read-only resolver v0.

Produce candidate links from existing data only.

Curate at least 50 ground-truth cases.

Measure precision and recall separately.

Publish examples and failure modes before score influence.