Trying to understand zeebe performance article

I’m working on an open source process engine written in Elixir, and I’m curious about the typical performance of a mature BPM platform. I came across this article. In the conclusion there is a little terminology that I don’t understand - “recorded latency”. What does that mean? Specifically, there are the statements:

  • 99% of requests are completed within the recorded latency (and)
  • 50% of requests are completed within the recorded latency

Later on in one of the “Release 8.x” sections,

I see “p99 is 2.5 seconds” and “p50 at around 0.500 seconds.”

I might understand this, but I"m not sure. Can you elaborate just a bit on what this means.

Charles