It's been a while since I looked at this code, but I'm probably the most familiar with it. The close loop benchmark you are looking at records each latency in a histogram called `recorder`. At the end of the benchmark, the latencies are added up and the 50th / 99th percentile is reported back. Also, the total number of RPCs made during the interval are sent back too. (based on the the number of items recorded to the histogram).
The setup for this is more complicated than we would ideally like it to be. The code you are looking was intended for you to see how we setup our code under test; the rest of the scaffolding is not as easy to follow sadly.