OptionalactualActual output from execution
User context
OptionalexpectedExpected output from test definition
OptionaljudgeLabeled, size-bounded evidence for judge oracles: the agent's final message, its final payload
and the artifacts the run produced. Set by drivers whose output is not all in the payload (an
agent that answers in its message, or one that leaves a manifest in its payload and the
substance in an artifact). Judge oracles read it in place of actualOutput when present;
actualOutput itself is unchanged, so structural oracles (schema, exact-match) are unaffected.
OptionalproviderOptional metadata provider to use for entity/SQL operations during oracle evaluation. If supplied, oracles should use this provider instead of the global Metadata.Provider to participate in the same transaction/connection as the calling test run.
OptionaltargetTarget entity (e.g., AgentRun)
The test being evaluated
OptionaltestThe test run being judged. The rubric oracle uses this as the subject record.
Oracle evaluation input