An Anthropic client instance from @anthropic-ai/sdk.
OptionalmodelClaude model to use for judging.
OptionalmaxMaximum tokens in the judge's response. Increase if reasoning is truncated.
OptionaltemperatureSampling temperature for the judge LLM.
Set to 0 for reproducible, deterministic evaluation runs.
Leave undefined to use the provider's default.
OptionalretriesNumber of retry attempts for transient errors (rate limits, 5xx). Uses exponential back-off: 500ms, 1s, 2s, ...
Configuration for the Anthropic (Claude) provider.
Uses structural typing for the client so you can pass any object that satisfies the interface — useful for mocking in tests.