Struct PairwiseScorer 

pub struct PairwiseScorer {
    pub model: Arc<dyn JudgeModel>,
    pub criteria_label: String,
}

Fields

model: Arc<dyn JudgeModel>

criteria_label: String

Implementations

impl PairwiseScorer

pub fn new( model: Arc<dyn JudgeModel>, criteria_label: impl Into<String>, ) -> Self

pub async fn compare( &self, prompt: &str, a: &Value, b: &Value, ) -> Result<(PairwiseChoice, String)>

Trait Implementations

impl AsyncScorer for PairwiseScorer

fn score<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, expected: &'life1 Value, actual: &'life2 Value, ) -> Pin<Box<dyn Future<Output = ScorerOutcome> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Treat expected as Response A and actual as Response B and run a pairwise judgment. The criterion label doubles as the task prompt context — sufficient for cases where the comparison criterion is fully described by criteria_label. Callers needing a richer prompt should use compare() directly.

Score mapping:

• A wins → score 1.0, passed=true
• tie → score 0.5, passed=true
• B wins → score 0.0, passed=false