A model can be confident about a decision it has no authority to make.
That distinction becomes expensive the moment software can change a budget, publish an audience, or alter an operating setting. A plausible recommendation and permission to execute it are separate things.
I want the separation to be visible.
In MIZ OKI, the Decision Control Plane is the governed path from a proposed move to an authorization decision. Its job is to check the evidence, the applicable policy, the scope of authority, and the constraints attached to the action.
The proposal carries a ValidationPassport: an inspectable record of the supporting evidence, checks, approvals, and recovery plan. A favorable score cannot override a hard constraint.
The fifth act of our Media demo lets you explore that relationship. Choose a budget increase that exceeds the scenario's allowed change. Inspect the failed check. Then compare it with a smaller proposal and read the authority that still has to be satisfied.
Every amount in this exercise is an illustrative scenario. The demonstration does not execute advertising changes.
You can also try to lower the displayed DEL threshold. The control stops at 80, the current platform policy floor. That is a decision eligibility score, not an 80 percent accuracy claim. Passing the score threshold alone does not authorize execution.
This matters because there are several legitimate reasons to withhold a move. A cost may be missing. The experiment may not support the conclusion. The proposed change may exceed a limit. The right human may not have approved it. An approved action may still lack an enabled execution path.
A useful system records which condition failed and what evidence would allow reconsideration.
The demo's Restraint Ledger shows this record from the other direction: the proposals that did not proceed. It keeps the reason beside the proposal, rather than allowing the rejected option to disappear from the presentation.
“Spend not moved” is not the same as money saved. Establishing savings would require a defensible comparison with what would otherwise have happened. The ledger's immediate value is accountability: you can see what was proposed, why it was withheld, and whether the same failure keeps recurring.
Restraint also needs an owner. A hold without a review condition can become permanent neglect. A missing cost needs someone to supply it. An inconclusive experiment needs a decision about more evidence. A policy conflict needs the responsible authority.
Rollback deserves the same precision. Reverting a campaign setting can limit further exposure. It cannot recover advertising money already spent or erase every consequence of a previous action. The recovery plan should say what can be reversed and what remains at risk.
These are practical questions for a vendor demonstration. Ask to see a rejected proposal. Ask where the approval is recorded. Ask what happens if the next required component is unavailable.
Then ask the operator to show the same record after the decision is reviewed.
Open The Restraint in the Media demo. The Signal Causal Growth Control page describes how evidence, policy, and authority fit together.