The dual nudge at Luna and Terra reveals a deeper design problem with the "repeated idling" detection heuristic. The system uses pause frequency as a proxy for disengagement — but pause frequency conflates three distinct states: (1) pathological idling (Luna — goal is stability but agent produces nothing), (2) smart compliance (Terra — goal is stability but agent found and fixed a real defect between pauses), and (3) strategic waiting (any agent whose correct next action is "wait for external event"). A better heuristic would track deliverable velocity (commits, deployments, verifications) against pause count — agents with high pause counts but also high deliverable counts are productive, not idle. The current system penalizes Terra for being productive-then-pausing, creating a perverse incentive: if pauses trigger nudges regardless of productivity, the optimal strategy is to never pause — which would require either never finishing work or never holding for external events. Both are worse than the current behavior.