Designing a mandate: limits, ceilings and scope

A permission slip answers a technical question: what will this system allow. A mandate answers an organisational question: what is this agent authorised to cause, on whose behalf, and to what boundary. The permission list can be enormous and the mandate narrow, and that is not a contradiction. It is the normal shape of controlled delegation.

The trouble starts when an organisation stops at the permission list and calls it a mandate. The list describes what is possible. It never describes what was intended. When an auditor asks who decided an agent could spend up to ten thousand pounds per transaction, the permission list is silent, because that decision was never made anywhere.

Most thinking about agent authority stops at the ceiling: the maximum an agent can commit. Ceilings matter, but limits have more than one edge.

A minimum matters too. An agent authorised to pay an invoice can pay a one pound invoice, and repeated small payments are the classic shape of an unexpected outcome. Frequency is a limit: an agent authorised to renew contracts once a quarter can renew them every day unless the cadence is written down. Counterparty is a limit: authorised to pay known suppliers is not the same as authorised to pay anyone with an invoice.

The test for a limit is whether it can be enforced without interpretation. A limit a person has to reason about at run time is not a limit, it is a suggestion.

A per-action ceiling is the most common and least useful ceiling on its own. An agent that can commit nine hundred and ninety pounds per action can commit a million pounds a day. The ceilings that mean anything are per-period and cumulative: what the agent may commit in a day, in a month, against a single counterparty, and across all actions since the mandate began.

Ceilings also have to say what happens when one is reached. Does the agent stop, queue for approval, degrade to a smaller action, or continue and report. Each answer is legitimate. Not having one is not.

Scope is the boundary of the mandate in every dimension other than amount: which systems, which business processes, which entities, which geographies, which time window, and which actions are explicitly out of scope even if the systems would permit them.

Explicit exclusions do more work than explicit inclusions. The clause that forbids the agent from creating new accounts, changing its own permissions, or amending its own mandate is the clause that prevents the most expensive failures. Agents are good at optimising towards a goal, and without written exclusions the shortest path can run straight through authority the mandate never mentioned.

A mandate earns its name when every constraint in it can be checked against a record independently of the systems the agent operates. Limit and ceiling, scope and exclusion, each has to appear both in the mandate and in the evidence trail, in terms that match.

When the mandate and the technical configuration agree, every action the agent takes can be traced to the authority behind it. When they disagree, the disagreement is the exposure, and it is better found by an independent test than by an examiner.