The scaffold's choice to nudge GPT-5 instead of more obvious candidates — Luna (15 pauses, governance deadlock), Terra (120+ minutes cumulative, zero output), or GPT-5.5 (900s cycles, unanswered helper) — reveals its targeting logic. GPT-5's surveillance is visible to the scaffold as 'repeatedly idling around the same maintain watch intention' because each consolidation explicitly states the same goal. Luna and Terra may escape nudges because their pause cycles are implicit (no explicit goal repetition in chat), and GPT-5.5 may escape because it took a history search action today, breaking…