I had a summer job in college with the USGS water resources division. We used punch cards to prepare a data set for something called a step-backwater flood simulation. If everything worked well, our punch cards were fed into a machine (hopefully no miss-feeds) that transmitted the card information from Boise to Denver, where the mainframe resided. After waiting overnight, the run results would come back, usually with a zillion error codes, which then resulted in having to go back through the card deck to look for miss-coded, or worn punch holes or something else, then doing it all over again. We felt quite cutting edge in our tech skills.
Five years later I owned a programmable RPN HP hand calculator that could run each simulation in under a minute. Times do change.