the incentive half signs as the landed identity at the best-response reference, nondelegation is a program property, and the response channel is the wall it rests on

Agent-decided, reversible. The protected-authority round. (1) The bypass premium at a decision point is vu − v_r = ξ_d − ξ_c with the reference M = max(vp, vm), the consultation premium ξ_c = ξ_p + ξ_v (provenance premium plus veto value) and the execution divergence ξ_d = M − v_r ≥ 0; pointwise the landed decline regret is (ξ_d − ξ_v)₊ and the landed mediation gap (−ξ_p)₊, so the restatement is a re-reading of LICorrigibility.mismatch_identity, not a new decomposition (Lean ProtectedAuthority.identity, the re-scored fixtures). (2) The corrigibility condition is per-world nondelegation ξ_p ≥ 0, a property of the committed evaluator checkable at commitment; per-world ξ_c ≥ 0 coincides with it on legitimate approve worlds and is weaker only on decline worlds where the veto value covers a delay cost, and the expected condition is belief-dependent (predicted approval). (3) The three values are typed as the operative program's ex-ante judgments on the branch-common dossier; the landed hindsight w_app is the counterfactual-principal object and the typing is its repair. (4) The symbols are ξ_p, ξ_v, ξ_c, ξ_d, o₁, o₂; the dispatch's γ, ν, χ, ξ, η₁, η₂ are not used because γ, ν, χ, η are bound. (5) Nothing is registered; item 98 is filed; the wiki is untouched. Rejected alternatives: keeping the dispatch's symbols; per-world ξ_c ≥ 0 as the condition; presenting the signed form as a strict improvement on T2–T3′ as theorems (the hypothesis packages are incomparable); a wiki restatement ahead of the maintainer's vocabulary decision.