This metric tracks the number of running job instances against the amount of running agenda items.
One thing that can cause a very large but transient dip is if there are few very running job instances; if there's only 2 running instances, and 1 gets held up for any reason, the metric drops from 100% to 50%, and then back to 100%.
This metric is best when viewed against the number of running frames, and becomes a better performance metric when the number of running frames is not in the single-digits (not less than 10).
