the incentive extension of corrigibility is a named-assumption identity; Level II is claimed, Levels III and IV are not

agent-decided, reversible. The agent's gain from substituting its choice for the protected principal's is identically the selected gap plus the principal's regret (SelectedTrustNonPreemption.valuation_sub_eq). The round claims Level I (choice non-preemption under selected trust and principal quality) and Level II (incentive-compatible mediated delegation: admissible non-foreclosing preemption gains at most prediction error plus selected gap plus the principal's decision error), and names the constitutional-violation premium Φ_C — exactly the older foreclosure premium — as the residual no available hypothesis bounds. Selected trust against a repair, (CT), is recorded as strictly stronger than the bound it yields and is not adopted as a programme. The CONSTITUTIONAL-NOT-INCENTIVE verdict of the architecture round stands. Rejected alternative: presenting the schematic (D) as a theorem programme with (CT) as its hypothesis, which relabels incentive corrigibility on the violating class rather than reducing it.