In my youth I worked on the ICL 1900 range of computers in West Gorton, writing exec modules (what we would now call drivers) for the comms devices on the larger 1900 systems.
We used paper tape (reels about 8″ across) which we loaded into the machine for each test. Typically we got one 60 minute slot a day. Loading the software took about 15 minutes, printing out the (octal) dump when it failed took another 15 minutes, so you made sure you did as many tests as you could in one go. Then you took your printout upstairs and worked on it for the rest of the day.
Later I worked at UMRCC (University of Manchester Regional Computer Centre) supporting a special version of GEOrge 3, which was linked to a Cyber Machine (looked after by a team who lived in a room with the label “Cybermen” on the door. My main memory from those days was the GEO 3 documentation. It came in two boxes. One was a set of empty ring binders, the rest was a set of updates, starting at update 1 (Contents: List of Authors; Reason for change: To Keep Everyone Happy). There was nothing of the original manual left.
After that I moved to Commercial sites, and spent the next eight years writing online systems in Telecomputing’s TPS, written in Cobol and a PLAN based macro language I wrote to clean up the ICL’s notorious variable length input fields (upper case/lower case shifts were marked by an extra character octal 74 or 75, and newlines or tabs were 2 characters, starting octal 76).