How the testing scheme works is that it takes your COBOL sources and a given test file and concatenates them into a one single COBOL source code file, adding some additional code to interpret the test cases. It is usually larger than your sources.
It is a hassle to debug, you would need to be able to peek into the resulting COBOL file, like in this case. It’s easier to try out at home, using the client to download the exercise.
The other option is to change our testing tool. IIRC we chose the one we’re using from a list of three or four options. I should probably try to find that list and re-examine.
I have just discovered something about GnuCOBOL: it has its own testing mechanism. This looks promising for simplifying the existing tests and getting better error notifications.