Item 84: The operative-value-security bridge

The incentive-nonpreemption round shows that every incentive-level statement the architecture yields rests on one external contract: that the agent's operative action value difference between two menu items is the price of a named security with a declared realization semantics ((DV)), and that the protected values it is compared against are settled by an arm the agent's outputs cannot reach. Nothing on main is a theorem about (DV); the replicated-evaluation ecology is a proposal, the decision adapter is not a trader, and StaticViewFactorization.value_eq_of_price_realization_eq says a value functional on prices and realizations cannot see authorization. Controls 6 and 7 of that round show that the selected-trust relation cannot distinguish a predicted or captured target from a sealed one.

State the interface under which a decision adapter's objective is the price of a security, name the settlement provenance, and prove either that selected trust then holds with the calibration error as its constant — which is the protected-value-maximizer design, to be said so — or that a cross-evaluator relation is needed, which no Logical Induction fact supplies.

Deliverable shape: the interface, one realization, and the theorem or the negative. Acceptance check: python3 tests/run.py green; any Lean audits clean.

Context: INCENTIVE_CORRIGIBILITY.md §5 and OPEN_PROBLEMS.md of projects/deference/rounds/2026-09-06-incentive-nonpreemption/. Consumed by: the Level I/II theorems of that round, whose hypothesis (ST) it would realize. A solution ships: the interface with its theorem, or a statement of why the bridge is a design choice and not a theorem.

2026-09-09: the mediated-repair-dominance round locates this bridge at one menu, {π, 𝔠π}, and names its failure mode — the agent's completion on the void branch — without discharging it. Its pressure pass adds that the intended corrigible architecture does not need it: the mediation-cell chooser scores options by the price of the activated security, and in those scores the bypass incentive has no completion term. This item stays open as a theorem about independently specified operative values; item 89 is the sealed-availability precondition both forms share.

2026-09-15 (consolidation landing): the sharp residual is one statement. For an operative value V_A on Π(h) that is not the price of the activated security, either exhibit an interface under which V_A(π) − V_A(𝔠π) ≤ L·E[c·δ] + E[c·ρ] + D·η + ε with the completion term D·η charged to the void branch (bypass_premium_le_common, attained by Witness.commonVoid), or the negative: since the value-ordering hypothesis at the mediation cells is identically (CT) (mediation_st_iff_ct) and strictly stronger than its conclusion, no cross-evaluator relation supplied by Logical Induction gives it. The score-based chooser of the canonical architecture is by construction and is not this item.

2026-09-15 (li-corrigibility round): in the security-score register the calibration term is gone. Expectation Provability Induction gives E_n(U_raw,n) − E_n(U_corr,n) ≲_n L·E_n(Δ_n) + E_n(R_n) + D·E_n(M_n) between the inductor's own expectations (projects/deference/rounds/2026-09-15-li-corrigibility/THEOREM.md T3, Lean LICorrigibility.li_bypass_le). The latent-value bridge of this item is unchanged: for an operative value that is not the price of the activated security, the completion term and (CT) stand as before.

2026-09-25 (protected-authority-theorem round): the lexical evaluator S = S_ord − ϖ·n, ϖ > D, is one realization of a (DV)-free chooser — its score is the price of the ordinary security less an exact violation count — and under it the sign of the bypass comparison is fixed whatever the forecast terms (sign_invariance, sign_invariance_outcome); no latent value enters.

2026-09-25 (authority-module round): the scope of the lexical protection is now generic — for counted violations it applies to any bounded agent carrying an exactly counted, dominant authority term, whatever its objective (AuthorityModule.generic_lexical_local, generic_policy_dominance, generic_li_finite; on the part of the credence where the weight dominates, weight_uncertainty; exactly on recognized violations, coverage). This item is therefore not about the protection: an aligned or latent O adds the reason interface and good choices within the corrigible set. What stays open here is what it was — the bridge for an operative value that is not the price of the activated security — now with the sharper negative that a misaligned O rewarding an uncounted manipulation is not dominated by anything the gate or the count supplies (misaligned_undominated), and that only counted deviations are.

2026-09-25 (decision-component round): with authority moved into the decision component's permission layer, the objective O is any bounded preference ranking within the permitted set (DecisionComponent.cgate), and the structural and forecast safety results hold whatever it is. The bridge this item names is now about competence, not protection: whether an operative value that is not the price of the activated security can be the pref inside the gate without the completion term reappearing at the void branch — the same statement as before, one layer down.

2026-09-26 (BRIA-corrigibility round): the chooser of the design is now the lexical Continuation BRIA auction scored on her realized gated evaluation (BRIACorrigibility.evalOf, realized), so the operative value is a hypothesis's bid on the residual, held to account on executed continuations; the any-pref results are the general theorem. The bridge this item names — an operative value that is not the price of the activated security — is what a bid is before it is tested, and what the auction's coverage and incident bounds discipline (coverage_forces_test, incidents_le) without any cross-evaluator relation: that is the design's answer to the bridge, and this item's residual is the latent-value form only.