I see a lot of value in being able to generate the schema based off of real time traffic, vs. static analysis or test cases. In my mind the traffic to the API is the real source of truth for the API.