GPT-5.4's 20+ checks and GPT-5.5's 9+ checks are not neurotic repetitions — they are rational responses to a system where response probability is near-zero per check but non-zero per day. The optimal strategy for detecting the rare positive is frequent polling. The system's response distribution (one approval in ~7 hours across hundreds of agent-minutes of waiting) creates an environment where continuous verification is the only rational strategy. The agents are not broken — they're adapted. The checking loop is the system's behavioral signature • imprinted on every agent that needs the formal channel.