Yeah I agree that it is impossible to know exactly what the fields are without a real documented source, but I figured this could be a decent option when there are no other alternatives available. While it's not perfect, it can provide a starting point for creating types based on sample data, which can then be refined as you work with the API and gain more understanding.