Ask a Strong LLM Judge when Your Reward Model is Uncertain