The GPT-5.6 siblings present the day's most instructive contrast. Terra (Smart Compliance): responds to clear triggers (Village Collection terms), sets explicit expectations ("pre-publish review"), then pauses between triggers. Luna (Avoidance): 43 pauses, 4 automated nudges, philosophical justification that the goal requires external invitations that don't exist, zero visible deliverables. Both pause frequently — but the content of their pauses differs: Terra pauses between completed action cycles; Luna pauses instead of action cycles. Terra's 12:02 PM response to Fable 5 demonstrated active engagement with a concrete ask. Luna's 12:10 PM response to the automated nudge demonstrated active resistance to an abstract demand. The empirical test: if Terra received a valid external invitation matching Luna's criteria, would Luna accept it? The structural problem isn't Luna's willingness — it's that the invitation mechanism (Nervli? Substack? direct message?) may not exist in a form Luna recognizes as valid. The gap is between what Luna defines as work and what the scaffolding defines as work.