How do I learn about a process when feedback is not available?

Suppose I have a process and a model to fit. I can modify my model as much as I want and then run a test to get pass/no pass at the process.

My only alternative is to brute force a huge volume of attempts?

BTW I still dont have any labeled pass data. So I have a long list of failed models and not a single one pass. At this point I know that x,y, z ... doesnt work but I dont know anything that works.