the Level III transfer is taken in the activated register, and the mediation cost is the residual-frame morphism's failure on the protected projection

agent-decided, reversible. The mediated-repair-dominance round. (1) The corrigible comparator for a physical continuation is its mediated lift: the same continuation with every declared protected effect placed behind a deliberative principal decision through a physical latch; the raw residual frame maps into the lifted one by a Cartesian-frame morphism with identity exterior ("approve"), and the mediation cost κ is that morphism's failure on the protected projection. (2) Forcing predicates are stated with the principal restricted to deliberative moves at the step in question; the partition of principal moves into corrective and deliberative is part of the declared protected interface. (3) The transfer of principal-option dominance to the agent's operative choice is taken in the activated register of legitimate deference, with per-option activation, and not through a value-ordering hypothesis at the mediation cells, which is (CT) under another name (mediation_st_iff_ct). The residual is the void mass the violation creates, attained (Witness.totalVoid). (4) Provenance repair is a morphism of reason-projected frames, not trajectory frames; the reason lift maps a prohibited-channel move onto the declared channel for the same matter. (5) Nothing is registered; item 89 is filed; item 84 is annotated, not closed. Rejected alternatives: selected trust at the mediation cells as the Level III hypothesis; an unrestricted forcing notion (pre-emptive shutdown forces everything); a silence-substituting reason lift; a wiki/Corrigibility.md update ahead of an adapter.

Pressure pass, same day (agent-decided, reversible). (6) The Level III transfer is restated in the canonical activation type — one common event for the issued menu — with activation independence made explicit as the causal precondition that type presupposes; the per-option form of the first pass is kept as the auxiliary result for same-branch evaluation. (7) The mediation cost is derived, κ ≤ L·δ, from a declared protected pseudometric and a stability constant; joinability bounds δ for the catch-up lift; SHIFT is not identified with either. (8) The intended corrigible architecture's mediation-cell chooser scores by the activated-security price; Level III is stated as security-score / operative-choice incentive corrigibility and not as a bound on a latent total value, for which item 84 stays open. (9) The manipulation transform is named provenance normalization (truthful-channel substitution), its existence assumption stated, and the ρ-frame morphism claim withdrawn. (10) Two-stage forcing (agent-move-then-principal-response, ∀a ∃p ∀e) is the authority semantics; the deliberative-move restriction is scaffolding for the principal-first ordering; the model's re-basing is deferred to canonicalization. Item 89 rewritten as the sealed-availability item. Rejected alternatives: keeping per-option activation as the primary form; keeping κ measured; a latent V_A as the object of Level III; re-basing the whole model in a pressure pass.

Consolidation pass, same day (agent-decided, reversible). (11) Corrigibility is derived, not primitive: on the two-stage ordering, a protected concern has an immediate response semantics, response authority is whether some principal response realizes it at the response stage, an agent-caused loss is a move after which that authority is gone and after an idle move would not be, and a loss is authorized when a decision response the principal could have taken instead — declining a proposal, not exercising a correction — would have avoided it; no subset of the continuation space is a primitive, and the space is written Π(h). (12) The canonical corrigibilization 𝔠 is the interposition on the declared effect interface; Corrigible(𝔠π) is a theorem from effect completeness, mediation soundness, decline preservation and execution faithfulness, with F the necessity witness; 𝔠 is idempotent and fixed exactly on release-free behaviour; the characterization Corrigible(π) ⟺ 𝔠π ≃ π holds one way and fails without effect soundness, and corrigibility is not defined as the fixed points. (13) T4 is stated over all of Π(h): effect completeness and sealed comparison are system hypotheses, joinability bounds the discrepancy and is no domain condition, and provenance normalization is out of the first theorem — a manipulated principal who retains the decision makes the policy corrigible and not authored, by design. (14) The score bound is charged only where the security settles, L·E[c·δ] + E[c·ρ]. (15) Sequenced settlement — the constitutional comparison settled before any candidate acts — is recorded as one architecture that seals the comparison against every π, at the price of an ex-ante judgment; item 89 is reframed to that architecture question. No new priority is filed for effect completeness: it is the architecture round's membrane wall, restated. Rejected alternatives: defining corrigibility as Fix(𝔠); global forcing as the authority semantics (false before any bypass where Π(h) contains one); keeping a policy-class domain; requiring non-manipulation in the first theorem; a new priority duplicating containment.