研究《Beyond QASecClaw: Agent Scaling, Verifier Feedback, and Evidence Reliability in Software Assurance》通过五个实验探讨了多智能体协作与验证反馈在软件保障中的可靠性边界。