pub struct LttResult {
pub alpha: f64,
pub delta: f64,
pub threshold: f64,
pub feasible: bool,
pub empirical_risk: f64,
pub n_served: usize,
pub false_accept_rate: Option<f64>,
pub diagnostics: Vec<LttDiagnostic>,
}Expand description
Result of LTT threshold calibration.
Fields§
§alpha: f64Target served-failure rate.
delta: f64Family-wise error rate bound (FWER ≤ δ over the qualified candidate grid).
threshold: f64Chosen threshold λ; serve iff score ≥ λ.
f64::INFINITY when no threshold is certified (infeasible, serve nothing).
feasible: boolWhether any threshold was certified. false ⇒ target infeasible on this data.
empirical_risk: f64Empirical served-failure rate at threshold (failures / n_served on the calibration set).
n_served: usizeServed-set size at threshold.
false_accept_rate: Option<f64>Gate empirical false-accept rate at threshold: fraction of INCORRECT calibration items
with score ≥ λ — the verifier ROC point the threshold is built on.
None when the calibration set contains no incorrect items.
diagnostics: Vec<LttDiagnostic>Per-λ diagnostics for every candidate in the grid, strictest to loosest.
Includes candidates below min_n (which appear with certified: false, p_value: 1.0).