Skip to content
AgentV

The Trust Protocol

Forensic Result Integrity

Section titled “Forensic Result Integrity”

For academic and industrial publications, it is not enough to report a score. You must provide a proof of integrity. AgentV implements the Trust Protocol, a cryptographic framework based on the Verification Certificate (VC) v3 standard.

ED25519 Forensic Signing

Section titled “ED25519 Forensic Signing”

Every evaluation trace in AgentV can be “sealed” using asymmetric cryptography (ED25519).

  • Evaluator Identity: The IdentityService manages your research identity, protecting your private signing key.
  • Trace Sealing: When a run is certified, the entire run.jsonl trace is hashed. This hash is signed by your identity, creating a non-repudiable link between your persona and the result.
  • Artifact Binding: The Trust Protocol doesn’t just sign the JSONL; it binds all sidecar artifacts (HTML reports, trajectory plots) to the same certificate.

Certification Workflow

Section titled “Certification Workflow”

1. Execute & Certify

Section titled “1. Execute & Certify”

After a successful evaluation run, use the certify command to generate the signature.

Terminal window
agentv certify --run-id <run_id>

2. The Verification Certificate (VC) v3

Section titled “2. The Verification Certificate (VC) v3”

The output is a run_manifest.json file containing:

  • signature: The Ed25519 signature of the evaluation.
  • evidence_ledger: A SHA3-256 manifest of every artifact generated during the run.
  • system_id: The public key of the evaluator who conducted the study.

Verification & Hard Gating

Section titled “Verification & Hard Gating”

Peer reviewers can verify your results using the agentv verify command.

Terminal window
# Verify the integrity of a results bundle
agentv verify --run-id <run_id>

The engine will return SIGNATURE_MATCH only if:

  1. The behavioral DNA trace has not been modified.
  2. The report artifacts in the evidence_ledger match their test-time hashes.
  3. The signature was produced by the claimed public key.

Use in Papers: Include your Public Key Fingerprint in your research paper’s “Data Availability” statement. This allow auditors to search global agent registers (where applicable) and verify your findings independently.