What you suggest here is the logical extension of how the various grading methods compute the score. Therefore, there is a lot of merit. (And, as you say, all the information to do this is readily available in the code.)
However, it is a moderate amount of work to implement all the cases, and I am not yet sure that is necessary.
My inclination right now is to invoke creative laziness, and just go with the simpler thing I have implemented. Later, if it turns out it would be significantly more useful to have the more complex analysis, we could do more work then.
(The response analysis is computed when the Statistics report is viewed. Therefore, there is not difficulty in changing the response analysis later if we decide to.)
I realise also that at the moment I am hampered by not having any data (because we are only just starting to adopt the ordering question type at the OU). If we had some example questions with lots of attempts, then we could try out the various proposed ways of analysing the responses, and see how useful the output looks, but I am not yet in a position to do that.
But, thank you for discussing this with me, it has been really helpful.