Struct StructuredJudge 

pub struct StructuredJudge { /* private fields */ }

Available on crate feature eval only.

Structured LLM judge that produces typed verdicts.

Tries function calling first (via response schema), then falls back to prompting for JSON output with a lenient parser.

Implementations

impl StructuredJudge

pub fn new(model: Arc<dyn Llm>) -> StructuredJudge

Create a new structured judge with default configuration.

pub fn with_config( model: Arc<dyn Llm>, config: StructuredJudgeConfig, ) -> StructuredJudge

Create a structured judge with custom configuration.

pub async fn judge( &self, expected: &str, actual: &str, criterion: &str, ) -> Result<StructuredVerdict, EvalError>

Judge a response against expected output with a specific criterion.

Tries function calling first, falls back to JSON extraction. On unparseable response, returns score 0.0 with parse error in reasoning.

pub async fn judge_with_rubric( &self, response: &str, context: &str, rubric: &JudgeRubric, ) -> Result<StructuredVerdict, EvalError>

Judge with a custom rubric.

Evaluates the response against the rubric’s scale points and produces a structured verdict.