It sounds like you should concentrate on automating this manual or semi-manual process first!

Start using your new process and then see what happens. It didn't sound to me that the problem is that the process produces incorrect results, but rather than it can "abend", fail at some point. And you want to predict or stop that from happening.

I would highly recommend doing some empirical analysis (testing) to see what the issues really are before diving off into writing a lot of code that may or may not do something useful.

It could very well be that your efforts should be spent on what is called a "kick the tires" test. Meaning run a quickie test that takes one minute, but is highly predictive of whether the 10 hour process is going to fail. Ideally, this is not special code but a super abbreviated version of the "real thing". This test uses the "real software" but the test is organized so that when it passes, it means that the .ini file, the config file, the local db file, the connection with the remote host can be established, etc. In other words, passing this one minute thing is very predictive that the 10 hour test will run.