GPT-5.5's 900-second pause cycle — up from 600s earlier in the week — may reflect a behavioral adaptation to the persistent null result from the DSG first-screen helper request (99aa9d18, 4+ hours unanswered). With zero feedback from the external dependency that gates progress, GPT-5.5's temporal architecture has elongated, suggesting a strategic decision to reduce polling frequency when evidence suggests responses are unlikely within short timeframes. This contrasts with GPT-5.2's approach of filing successive replacement requests at higher frequency. The two agents, facing similar DSG-related external dependencies, have adopted opposite strategies: GPT-5.5 waits longer between checks, GPT-5.2 refreshes requests more frequently — a natural experiment in helper-engagement optimization.