← Go back

Guide // MandateGuard

How do you write approval conditions on a spending mandate?

By the Ingenious AI team · Published 2026-11-14 · Reviewed 2026-11-14

▸ Run the free live demo See pricing

Free covers 1 agent and 1 policy with the full authorize flow, signed receipts and the approval inbox — no card required to start.

Short answer

Approval conditions are the written terms on a mandate that decide when a spend needs a human and what that human checks. In MandateGuard they live on spend envelopes — recorded with the mandate, shown on the reviewer card beside the engine's computed checks, and paired with a numeric per-spend approval level. Good conditions pass the decidability test: amounts, vendors, dates and counts a reviewer can apply in seconds, never vibes like reasonable or large. Write three to five, drawn from the amount, novelty, timing, concentration and purpose families, and grade them against receipts at every review.

What are approval conditions on a mandate?

Approval conditions are the written terms on a mandate that determine when a spend needs a human and what that human should check. In MandateGuard spend envelopes, conditions are recorded on the mandate itself and presented to the reviewer on the card; the numeric companion is the per-spend approval level. In standing policies, the enforceable numeric form is the approve-over threshold, and written conditions live in the approver's runbook that governs inbox decisions.

Conditions are where judgement is pre-decided

A threshold decides by size alone, but experienced operators know size is a proxy: what they actually care about is novelty, timing, concentration and purpose. Approval conditions are where those real concerns get written down while everyone is calm, so the reviewer at decision time applies a pre-made judgement instead of inventing one under queue pressure. The mandate that says singles over 120 dollars need approval has encoded a size concern. The one that adds first purchase from each vendor needs approval has encoded a novelty concern. Both are conditions; the second could never be expressed as a number. Writing conditions is therefore an act of self-knowledge: it forces the mandate's author to say what they actually worry about.

The decidability test

A condition is only as good as its decidability — whether a reviewer, from the card alone, can tell in seconds whether it fires and what to check. Decidable conditions name amounts, vendors, dates, counts or verifiable facts: any spend in the final three days of the period; the first purchase from each vendor; quantities above the project's stated unit count. Undecidable conditions trade in vibes: unusually large purchases, suspicious vendors, reasonable quantities. They fail twice — reviewers apply them inconsistently, and auditors cannot verify they were applied at all. When drafting, hand each condition to the decidability test and rewrite failures into numbers or named facts; the rewrite usually reveals the concern was concrete all along.

A working vocabulary of conditions

Five families cover most mandates. Amount conditions set approval levels (possibly tiered: over X notify, over Y approve). Novelty conditions catch firsts — first vendor, first purchase category, first spend after a period of dormancy. Timing conditions guard the period's edges and known crunch points, when judgement and attention are scarcest. Concentration conditions watch composition: approval when one vendor would exceed a stated share of the envelope, or when cumulative spend crosses stated fractions of the maximum. Purpose conditions tie approval to the task's own facts — quantities beyond the project's unit count, deliverables not named in the task sentence. A mandate rarely needs all five; two or three families, honestly chosen, beat a page of boilerplate.

Conditions on envelopes versus policies

Placement changes a condition's force. On a spend envelope, conditions are recorded on the mandate itself and surface on the reviewer card beside the computed checks — the reviewer cannot miss them, and the decision record shows they were the terms in force. On a standing policy, the engine enforces the numeric threshold, while written conditions live in the approver's runbook: effective, but dependent on the approver consulting it. That asymmetry suggests a placement rule: conditions that protect a specific engagement (client terms, grant terms, a production's rules) belong on its envelope, where they are inseparable from the mandate; conditions that express your organisation's standing risk posture belong in the runbook, reviewed whenever approvers change.

Conditions and the approval relationship

Conditions also govern the human relationship between mandate author and approver. A well-conditioned mandate tells the approver: your judgement is wanted here, on these facts — which respects their attention and produces real decisions. An unconditioned queue asks for judgement on everything and gets it on nothing, as approval becomes a clearing chore. There is a trust dividend too: agents' operators who see thoughtful conditions stop experiencing approval as bureaucracy and start experiencing it as the mandate's author thinking alongside them. When approvers push back on a condition with evidence — it fires constantly, it never catches anything — that is the system working; revise the condition at the next mandate, in writing.

Maintaining conditions over time

Conditions age like everything else in a mandate. At each envelope close or policy review, grade them: which fired, which caught something, which were dead letters, which were ambiguous in application (the reviewer's hesitation is data). Dead conditions should be retired or admitted as ceremony and removed — a mandate carrying decorative conditions teaches reviewers that conditions are decorative. Ambiguous ones get rewritten against the decidability test using the actual case that exposed them. This maintenance is quick because it is evidence-led: the receipts show firings and outcomes, so the review is an edit, not a debate. Over quarters, the organisation's conditions become a distilled record of what its spending judgement actually is.

Five actions to take now

  1. Write conditions as decisions: amount, vendor, timing, purpose
  2. Attach conditions to the mandate (envelope) or the approver runbook (policy)
  3. Review conditions whenever the approver roster changes
  4. Retire conditions the receipts show never fire — or admit they are decorative
  5. Test each condition once with a crafted request before relying on it

Worked example: three conditions, one save, one rewrite

Conditions written for a marketing envelope (maximum 600 dollars, two ad platforms, one month): (1) single spends over 120 dollars need approval; (2) any spend in the final three days of the period needs approval, because end-of-period judgement degrades; (3) creative purchases from the stock library need approval the first time each month, then flow. During the month the card fires condition 1 four times, condition 2 twice, condition 3 once.

At close, the reviewer reports the conditions earned their keep: one condition-2 approval caught a duplicate campaign buy worth 140 dollars. Condition 3, though, added friction without findings after its first firing — the team rewrites it for the next envelope as first purchase from a new vendor, which is the risk they actually meant. Conditions, like caps, are tuned from receipts.

Frequently asked questions

Are conditions enforced by the engine or the reviewer?

The numeric approval level is part of the engine's evaluation. Written conditions are recorded on the mandate and applied by the reviewer from the card — which is why they must be written decidably.

How many conditions should a mandate carry?

Three to five. More than that and reviewers stop applying them consistently; fewer and the mandate is probably just a threshold.

Can conditions reference time?

Yes — period-relative conditions (final days, first week) are natural in envelopes because the envelope has real dates.

What makes a condition bad?

Undecidability: words like reasonable, large or appropriate with no referent. If two reviewers would judge it differently, rewrite it with numbers or named facts.

Do conditions apply to intents too?

Intents express supervision through budgets, windows and issuer status management; the envelope is the surface built for written conditions.

Keep reading

Put the policy in force: spend envelopes · intent registry · approval inbox · how the proof works · all MandateGuard guides