pub struct EvalReport {Show 17 fields
pub clean: usize,
pub p_at_1: f64,
pub hit_at_k: f64,
pub mrr: f64,
pub relation_evidence_recall: Option<f64>,
pub relation_evidence_expected: usize,
pub relation_evidence_missing: usize,
pub confidence_expectation_rate: Option<f64>,
pub confidence_expected: usize,
pub confidence_missing: usize,
pub garbage: usize,
pub garbage_rejected: Option<f64>,
pub answer_precision: Option<f64>,
pub answer_count: usize,
pub by_class: BTreeMap<String, ClassStat>,
pub calibration: BTreeMap<String, ClassStat>,
pub cases: Vec<CaseResult>,
}Expand description
Aggregate metrics. Fractions are in [0.0, 1.0]; nullable fractions were not measured.
Fields§
§clean: usize§p_at_1: f64§hit_at_k: f64§mrr: f64§relation_evidence_recall: Option<f64>Fraction of declared route_relation_must expectations present on the top route hit.
None means the suite did not judge route relation evidence.
relation_evidence_expected: usize§relation_evidence_missing: usize§confidence_expectation_rate: Option<f64>Fraction of clean cases with a declared confidence expectation whose top confidence met it.
None means the suite did not judge confidence bands.
confidence_expected: usize§confidence_missing: usize§garbage: usize§garbage_rejected: Option<f64>§answer_precision: Option<f64>Precision of answer-grade top hits (exact/strong) across clean and garbage cases.
Clean cases count correct only at rank 1; garbage cases are correct only if not answer-grade.
answer_count: usize§by_class: BTreeMap<String, ClassStat>Per-class ok/total (sorted), so a large stratified set shows which class is weak.
calibration: BTreeMap<String, ClassStat>Per top-hit confidence band ok/total over CLEAN cases — does “strong” predict correct? The safety property for a trusted brain: strong should stay high-accuracy.
cases: Vec<CaseResult>Trait Implementations§
Source§impl Clone for EvalReport
impl Clone for EvalReport
Source§fn clone(&self) -> EvalReport
fn clone(&self) -> EvalReport
1.0.0 (const: unstable) · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
source. Read more