Test the contract.
Then test the agent.
A snapshot of an A2A runtime: discovery, AgentCard validity, a real message call, identity, signatures, security, and reliability history when monitored.
Canonical and observed AgentCard locations.
A2A schema, v1 interfaces, skills, and security.
Negotiated JSON-RPC message/send response.
JWS, provider identity, latency, and history.
What the score covers
Only what the scan can observe. “Not measured” is not treated as failure. Repeated monitoring adds availability, drift, stable latency and identity verification. Read the public methodology →