Clips measured.
Median fixed pipeline latency.
Median queue backlog.
95th-percentile queue backlog.
Median total lag behind the source.
95th-percentile total lag behind the source.
Worst total lag observed.
Median per-segment expansion ratio, or null when no segment had a source span.
Total spoken audio over total source speech. This is the drift slope: above 1.0 the backlog grows for as long as the speaker keeps going. Null when nothing qualified.
Seconds of lag added per minute of continuous speech, derived from the slope.
Aggregate view over every segment recorded for one language.