Behavior of models to adhere to natural‑language instructions reliably.