Kind of disqualifies you... but if you are getting onto Mainframes soon
Well, there's the Computer Science approach to a project. Chose the current fashionable languages (and frameworks, and all the other five "technologies" you'll likely need) and Immediately start looking for libraries, or instructions which happen to do exactly what you want. Slap some code together. Run it. Fix-up the places in the code which seem to cause the errors, then fix-up the next iteration which has occurred from the latest code (there will be fewer problems each time, so it'll only likely take four or five iterations of this process). Remember, if there are things you can't fix, give some thought to calling them "edge cases" or even "corner cases". If you can do that successfully, you don't need to bother about it.
There you go, job done. By this time, you'll wish never to see that code again, and you probably won't need to, and no-one but you marker will ever have to look at it, and no-one will ever have to change it. After a few weeks, it'll never even be run again.
OK, can't go that route because you are forced to do it in tacky old COBOL. Since you can't go the CS route - abandon everything associated with that.
What that means is, you design the program before wondering how to write the actual code. Use pseudo-code to set out how you think the processes will work. The great thing about pseudo-code is that you can write anything you like. And then, with your specification, you go through the code to make sure you are doing everything in the specification and doing everything correctly.
You can start at a high level, and once happy with that, get a bit lower. Once you are happy with the pseudo-code, then you start to think about writing the COBOL.
If, when testing your program, you find you missed something in the design, don't just patch it up, but step back to the design and fix that.
You'll end up with a program you are happy with, that you don't mind people looking at, that other people will be able to change, and which could happily be used for 20 years.
OK, don't expect it all at once, but you'll get there much quicker this way