Usually you should move-construct a StructArray from inputs[0] and index into it to get a Struct element, which can be then indexed with a fieldname as a C++ string to get a generic Array, with which you can then move-construct a TypedArray of correct type (get its real runtime type with Array::GetType if you can't ensure it at coding time). Finally you should usually index into that TypedArray to get a numeric element, get an iterator like cbegin/begin, or release it as a buffer_ptr_t to get a C++ native pointer.
void Example(matlab::mex::ArgumentList outputs, matlab::mex::ArgumentList inputs)
{
matlab::data::TypedArray<double> A = matlab::data::StructArray(matlab::data::StructArray(std::move(inputs[0]))[0]["test3"].operator matlab::data::Array())[0]["test4"];
A.cbegin();//Get an iterator
A[0];//Get a double element
const matlab::data::buffer_ptr_t<double> B = A.release();//After this, A becomes invalid
B.get();//Get a native pointer
}
20241028：Edited for forgetting to provide a type to the buffer_ptr_t
