Back in the late 80's I was assigned to the Holiday Inn support team as an IBM Program Support Rep. My job was maintenance of IBM software on their multiple systems, which included the T.P.F. based HOLIDEX reservation system. I had come from 18 years of hardware service, so my strength was a better than average understanding of the hardware/software interface, and knowledge of machine language at the op-code level.
But the TPF specialists worked at this level all the time. I finally decided that the key characteristics of TPF programmers were a thorough knowledge of 370 assembler, combined with an inability to read a clock, because they all seemed to work crazy hours and shifts - for which they earned about 50% above the usual salaries.
My only contribution to TPF was to fix a failure in its SADUMP program, that occurred on a 3083 mainframe with exactly 16384 kilobytes of main storage. It failed to handle the (expected) error on the 2k page instruction at the upper boundary and got into a loop. By that time, most mainframes either had more storage or used a little of the upper end of main storage internally, and shifted the top end down away from 16384k, and so did not have the problem. And VM hid the problem if it was running in a virtual machine.
Holding in there for 31 years was pretty good, I managed 28 . I was a Tech Support Rep in the marketing division so we really were not that technical, although I did have to learn some level of detail when working with VM.