It makes sense. I think one of the challenges is the question of how much you need to store in order to be able to do this. Hanging onto every request/response

It makes sense. I think one of the challenges is the question of how much you need to store in order to be able to do this. Hanging onto every request/response object / parameters for an interaction with an API can be become expensive pretty quickly. This becomes an issue when you start thinking about what is considered to be a bug. If you're just talking about error codes, then you're probably fine to store everything. But when you talk about unexplained (but not erroneous) behaviour from the API, how much you need to store to catch the bugs is an important consideration.