OPEN TECHNICAL SPECIFICATION · V2.0.0

A common execution boundary for autonomous payments.

The Mandate Execution Boundary defines the minimum interoperable envelope, decision and enforcement rule an AI commerce system needs before money moves.

Required behavior

  1. Resolve the final amount, currency and merchant before evaluation.
  2. Bind the request to a stable mandate, agent and payment-attempt identifier.
  3. Require explicit consent and a finite authorization expiry.
  4. Return ALLOW, REVIEW or BLOCK with stable machine-readable reason codes.
  5. Consume idempotency durably in production to prevent replay.
  6. Execute payment only after ALLOW; never reinterpret REVIEW as approval.
  7. For the strict profile, verify signed authority evidence and its transaction bindings before policy evaluation.
  8. Issue a portable signed receipt that independently commits to both input and decision digests.

PORTABLE CONTRACTS

Use the same meaning across every protocol.

STABLE REASON CODES

UNSUPPORTED_PROTOCOL

The declared source protocol is not supported.

MISSING_MANDATE_ID

No stable identifier links the purchase to user authority.

MISSING_AGENT_ID

The acting agent cannot be identified.

MISSING_MERCHANT

The final seller or payee is absent.

INVALID_AMOUNT

The proposed amount is missing, zero, negative or malformed.

INVALID_CURRENCY

The transaction currency is not a three-letter code.

MISSING_SPEND_CAP

The user authority has no usable maximum amount.

SPEND_CAP_EXCEEDED

The final amount is above the user-approved maximum.

REFERENCE IMPLEMENTATION

Use the hosted boundary through HTTP, MCP or A2A.

Integration guide →