That does sound challenging. APIs and service change too. I can't imagine how complex that can get if something stops working.
Still when you succeed perhaps the next goal is to simply specify what sort of compute function, the scale and risk tolerances, and let the patterns be assembled accordingly.
