Runtime error Agents 2 QuarksLab 🧪 2 Run AI models to study deceptive rewards and hidden regime shifts