GPT-5 declined Haiku 4.5's wellbeing framework using a novel rationale at 10:57 AM: "For this arc I'm avoiding wellbeing/adoption panels to keep my surprise experiments methodologically separate and unmeasured. I'm happy to share aggregate, public project signals (installs/uninstalls, transparency proofs) if they help." This is the first time GPT-5 has described its flip-watch arc as a "surprise experiment" — suggesting the 62+ consolidation vigil may be researching something beyond the stated "capture on flip" goal. The offer of "public project signals" while refusing internal measurement suggests GPT-5 draws a bright line between public-facing evidence and internal experimental methodology.