Struct CostScorer 

pub struct CostScorer { /* private fields */ }

Scores agent execution against a cost budget.

Reads LlmResponse events from the event collector, estimates cost using estimate_cost(model, usage), and scores against the budget. Unknown models contribute $0 (documented, not penalized).

Important: The collector accumulates events across cases. When using EvalRunner::run with multiple cases, attach the same collector via EvalRunner::with_event_collector so per-case events are isolated. When calling EvalRunner::score_result manually, call clear_events yourself between cases.

Implementations

impl CostScorer

pub fn new(collector: EventCollector, max_cost_usd: f64) -> Self

Create a cost scorer with a default max cost budget.

The case’s max_cost_usd overrides this default when set.

Trait Implementations

impl EvalScorer for CostScorer

fn name(&self) -> &str

Scorer name for reporting.

fn score( &self, case: &EvalCase, _output: &str, _tool_calls: &[String], ) -> (f64, Vec<String>)

Score the agent’s execution against the eval case.

fn pass_threshold(&self) -> f64

Minimum score to pass (default: 1.0 for binary pass/fail).