Design the Guardrail Layer for the Notification Policy

Product gives you three requirements for the notification RL system: (a) never more than 3 sends per user per day, no exceptions; (b) never send during a user's local quiet hours (10pm–8am); (c) "try not to over-message users who seem to be getting annoyed, but use judgment — sometimes a nudge is worth the small risk."

1. For each of the three requirements, state whether it should be implemented as a structural guardrail or as a term in the reward function, and justify the choice.
2. For the guardrail(s) you chose, specify exactly where in the request path (relative to the policy scoring step) each is enforced, and why that placement is correct.
3. Requirement (c) is deliberately fuzzy. Design a concrete mechanism for it that respects the "use judgment" intent without becoming a loophole that swallows requirement (a).