Here is my limited knowledge (based primarily on observation) of the various thread stats:
This is the measurement of time between the receipt of inbound traffic. So you receive the first message since zeroing out the stats. The timer starts. When you receive the next message, the timer stops while the message is processed on the inbound side and then starts again once the message is dispositioned. The count is cumulative and only resets when the stats get zerod out.
This is a measurement regardless of whether this is an Inbound or Outbound thread (each thread has an inbound and outbound side remember). So obviously the value of the data will be different depending on the orientation of the thread (Inbound or Outbound).
There are other considerations when trying to make sense of this data.
This is the measurement of the time elapsed between sending of messages. Just as with IB Latency, this value is measured for both Inbound and Outbound threads and has the same caveats as IB Latency.
Again there are other considerations when trying to make sense of this data.
At first blush one would think this should be the sum of IB Latency and OB Latency and sometimes it is. However, this value depends on the configuration. Rarely is this number exactly calculable from the displayed data by humans. I suspect therefor there is some additional rounding or truncation going on behind the scenes.
Also there is no figure presented that gives end-to-end time. That needs to be calculated by you and is extrememly dependant on local knowledge of the integrations.
I once had a discussion with someone in the old HCI days (Streepy, etal) about these stats and gained some insight. However, I do not recall ever seeing any written explanation of how the stats are collected or computed and after the Bob Murray purge of the developers, I am not sure there is anyone left at Quovadx who really understands how these stats are derived.
What I have given here is just a small description. This topic can get much more complex. I suspect this is one of the reasons that although I have provided a proc to completely dump the msi stats in a delimited format for import into statistical analysis tools to a number of people, I have not heard of a single organization making use of the stats beyond number of messages, etc. When attempting to make sense of the data many just give up.
I have to admit, I have never been given sufficient time by any of my employers to properly make use of these statistics. Dumping the stats is easy, crunching those numbers in a meaningful manner is not.
There is a lot more I can discuss with you.
Email me if you would like to chat in more depth.