In long task chains for LLM Agents, meaningful reward signals may only appear upon final task completion, making it difficult to correctly attribute credit to intermediate steps—a phenomenon in reinforcement learning known as the "credit assignment problem."