Tag
OpenAI discusses how simulated deployments reduce evaluation awareness to near real production levels, and extends the method to agentic deployments with stateful tools using tool simulators.