Authorization grants bounded permission. Execution is a separate operational event. A governed financial system should not collapse permission into action.
An authorization is a decision about what may happen. It identifies a permitted action, the authority that granted permission, the scope of that permission, and the conditions under which it remains valid. Execution is what happens when an operational component attempts to turn that permission into an external consequence.
Those states are related, but they are not interchangeable. Between them, time passes, evidence can change, controls can produce a different result, external systems can become unavailable, and authority can be revoked. Treating authorization as if it were already execution erases this interval and the responsibilities that belong inside it.
Permission has a scope
A useful authorization should be bounded. It may specify an instrument, amount, destination, time window, permitted venue, risk context, or other explicit conditions. It should also be attributable: the system must be able to identify who or what held the authority to permit progression.
Authorization does not mean that every future attempt remains valid. A market state may change. Evidence may become stale. A limit may be consumed by another action. A mandate may expire or be withdrawn. The permission must therefore be evaluated in relation to current state at the execution boundary.
A separate boundary preserves control
A distinct execution boundary gives the system a final place to verify that the instruction still corresponds to the authorization, required controls still pass, the destination is permitted, and the authorization has not expired or been revoked. This is not a second discretionary decision by the execution component. It is a deterministic verification that the operational event remains within the granted permission.
The separation also supports failure handling. An authorized action may never be attempted. An attempt may be rejected before transmission. A transmission may time out. An external system may acknowledge receipt without completing the requested change. Each outcome has a different operational meaning and should produce a different state.
Revocation remains meaningful
If permission and action are collapsed, revocation can become performative: by the time the authorization exists, the consequence may already be underway. A separate execution state creates a clear interval in which permissions can expire, be withdrawn, or be superseded before an external instruction is sent.
This improves more than control. It improves auditability and attribution. Investigators can distinguish who authorized an action, which component attempted it, what evidence existed at each point, and whether the final instruction remained within scope.
The record should reflect the distinction
A governed record should preserve at least the authorization identity, scope, authority, effective period, relevant evidence and control state; then separately record the execution instruction, attempt time, destination, transmitted representation, acknowledgements, failures, and observed outcome.
Keeping these states separate does not make intelligence less useful. It creates the conditions under which intelligent proposals can progress without acquiring invisible operational authority. This extends the principle established in Why governed AI needs execution boundaries: capability may inform a decision, while the right and mechanism to act remain explicit.