pub struct CalibrationRecord {
pub actor: String,
pub n_predictions: usize,
pub n_resolved: usize,
pub n_hit: usize,
pub n_expired: usize,
pub hit_rate: Option<f64>,
pub brier_score: Option<f64>,
pub log_score: Option<f64>,
pub reliability_buckets: Vec<(f64, f64, usize)>,
}Expand description
Per-actor calibration summary computed over the resolved subset of the actor’s predictions.
Fields§
§actor: StringStable actor id (e.g. reviewer:will-blair).
n_predictions: usizeTotal predictions made by this actor in the frontier.
n_resolved: usizePredictions that have been resolved (have an associated
Resolution). Open predictions don’t contribute to scoring.
n_hit: usizeResolved predictions whose matched_expected = true.
n_expired: usizev0.40.1: predictions closed by the calibration runtime
without an explicit Resolution (deadline passed). Counted
separately from n_resolved so the predictor still answers
for the missing commitment without their Brier or log score
being moved by it.
hit_rate: Option<f64>Hit rate over resolved (or None if n_resolved == 0).
brier_score: Option<f64>Brier score, lower is better. None if no resolutions.
log_score: Option<f64>Log score, higher (closer to 0) is better. None if no resolutions.
reliability_buckets: Vec<(f64, f64, usize)>Bucketed reliability diagram: for each predicted-confidence
band, the observed match rate. Empty bands are omitted.
Format: (confidence_lower_bound, observed_hit_rate, n_in_band).
Trait Implementations§
Source§impl Clone for CalibrationRecord
impl Clone for CalibrationRecord
Source§fn clone(&self) -> CalibrationRecord
fn clone(&self) -> CalibrationRecord
1.0.0 (const: unstable) · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
source. Read more