pub struct DriftBaseline {
pub completion_rate: f64,
pub dead_letter_rate: f64,
pub latency_p95_ms: Option<u64>,
}Expand description
The promotion-time baseline drift is measured against: the metrics the promoted version’s own evaluation recorded for the arm it displaced (the floor’s recorded baseline — “is the promoted version regressing against the evidence that promoted it”, nothing deeper).
Fields§
§completion_rate: f64Baseline completion rate, in [0, 1].
dead_letter_rate: f64Baseline dead-letter rate, in [0, 1].
latency_p95_ms: Option<u64>Baseline p95 decision latency (the journaled
dependency_latency_ms feature), when the evidence carried one.
Implementations§
Source§impl DriftBaseline
impl DriftBaseline
Sourcepub fn from_twin_report(report: &Value) -> Option<Self>
pub fn from_twin_report(report: &Value) -> Option<Self>
Read the baseline out of a serialized twin evaluation report — the
aggregate section TwinCandidateEvaluator writes. None when
the report names no such section (a non-twin evaluation, a
hand-authored report): drift detection requires the promotion-time
evidence, and it fails absent rather than guessing a baseline.