Strategy vs authority

An AI agent can decide what it wants to do. Grantline decides whether the active authority permits it.

That distinction separates three stages:

1. The agent forms an intention.
2. The agent turns that intention into an Action Plan under a Mandate.
3. Grantline evaluates the plan before the controlled execution path can act.

Grantline does not decide whether the agent's strategy is profitable or sensible. It checks whether the proposal fits the authority that an owner or another authorised agent granted, together with the conditions attached to it and the current state relevant to the action.

Authorisation is not execution

ALLOW, ESCALATE, and DENY are authorisation decisions, not execution results.

• ALLOW means the proposal passed the current checks and may enter the authorised execution path.
• ESCALATE means the proposal requires additional approval before execution.
• DENY means the proposal cannot proceed through the authorised path.

An allowed action can still fail when the downstream execution system rejects it. That failure is an execution result, not a change to the earlier authorisation decision.

The authority boundary

The agent proposes an action, but it does not own the controlled capital. Grantline checks the proposal against the authority granted to that agent, and only a proposal that satisfies those boundaries can reach the execution path. The Vault remains the custody boundary, so a signing key does not become an unrestricted capital-movement key.

See Decisions for how Grantline communicates the result of that check.