OptionalexecutionScope: AIPromptExecutionScopeThe evaluating run's configuration, runtime API keys and credential scope.
An agent's self-check passes its own, so judging a customer's run spends the customer's keys —
and under 'RuntimeOnly' never the platform's. Omitted, the judge resolves keys as any prompt does.
A RubricEngine whose catalog, evaluations, prompts, decisions, and agent runs use the caller's provider.