koog-agents-optimization/ai.koog.agents.optimization.training.metrics.impl/ElapsedTimeMetric

ElapsedTimeMetric

@Serializable

@SerialName(value = "elapsedTime")

data class ElapsedTimeMetric(val key: MetricKey<ElapsedTimeMetric> = KEY, val avgPerFinishedSubstage: Duration? = null, val maxPerFinishedSubstage: Duration? = null, val etc: Duration? = null, val withRetries: ElapsedWithRetries? = null) : Metric

Derived timing aggregates for a stage.

The stage's own wall-clock duration lives on StageRecord.realElapsed — this metric only exposes values that are genuinely derived from substages or from the stage's own elapsed:

avgPerFinishedSubstage: mean elapsed time across finished substages. In-progress substages are excluded — including their partial elapsed would skew the average.
maxPerFinishedSubstage: longest finished substage. Useful for spotting outlier items in a dataset iteration.
etc: projected time-to-completion. Prefers the running direct child's own ETC when available (plus projected time for siblings not yet started) — this covers the common "single iterateDataset child" case that the naive realElapsed / processed × remaining formula can't handle. Falls back to that formula when no child ETC exists.

Constructors

ElapsedTimeMetric

constructor(key: MetricKey<ElapsedTimeMetric> = KEY, avgPerFinishedSubstage: Duration? = null, maxPerFinishedSubstage: Duration? = null, etc: Duration? = null, withRetries: ElapsedWithRetries? = null)