Section C• Last verified November 2026
Spending Rules & Approval Requirements
Deterministic policy verdicts: ALLOW, ALLOW_WITH_LIMITS, REQUIRE_APPROVAL, ESCALATE, DENY.
Deterministic Policy Outcomes
Whenever an agent proposes a transfer, the Universal Policy Engine processes the parameters against the active mandate and returns exactly one of five deterministic verdicts:
Verdict Reference Table
| Policy Outcome | System Behavior | Requires Human? |
|---|---|---|
| ALLOW | Execute autonomously; within all limits and thresholds | NO |
| ALLOW_WITH_LIMITS | Execute within modified or clamped parameters | NO |
| REQUIRE_APPROVAL | Escrow in pending state; await human wallet signature | YES |
| ESCALATE | Halt automated task; flag for operational supervisor review | YES |
| DENY | Hard rejection; prohibited recipient, exhausted budget, or expired mandate | BLOCKED |
When is Human Approval Required?
Human review is strictly reserved for high-value or ambiguous actions:
- The proposed transfer amount exceeds the mandate's
approval_threshold. - The recipient address is valid but has not yet been explicitly whitelisted.
- The DAG plan was mutated or rescheduled during active execution.
Fail-Closed Principle
Fail-Closed Architecture
If any policy rule encounters a database timeout, missing signature, or ambiguous condition, the policy engine automatically defaults to DENY. The system NEVER “fails open”.