Reduce model quota spent on repeated supervision, polling, context loading, and status chatter when the user asks for quota-saving operation. Preserve scientific rigor and task quality; do not silently downgrade models or narrow the goal.
