Skip to main content

compute_scores

forge_engine::lab::evaluate

Function compute_scores 

Source 
pub fn compute_scores(
    check: &CheckResult,
    patch: &StructuredPatch,
    task: &EvalTask,
    store: &ForgeStore,
    question_sig: &str,
    config: &ForgeConfig,
) -> ForgeResult<ScoreVector>
Expand description

Compute full score vector.

correctness, novelty, and stability are raw metrics in [0.0, 1.0]. weighted_total applies the task weights: w_c * correctness + w_n * novelty + w_s * stability. Gates (correctness_gate, promotion_min_suite_pass_rate) apply to the raw metrics.