Guide // MandateGuard
What is vendor trust scoring from your own receipt history?
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
Vendor trust in MandateGuard is a score computed only from your own workspace's receipt history — the allowed and denied decisions and totals your agents have actually accumulated with each vendor, shown beside the score. It uses no external ratings and no other customer's data, on principle: the question that matters is not whether a vendor is reputable in general, but how it has behaved with your agents under your mandate. Repeated over-cap attempts, duplicate escalations and clean boring histories all become visible per vendor, which makes trust a practical tool for allowlist reviews and renewals rather than a decorative badge.
What is a vendor trust score?
Vendor trust in MandateGuard is a per-vendor summary computed only from your workspace's receipt history: how many decisions were allowed and denied, and the total allowed value, presented beside the score so the evidence is never separated from the verdict. It deliberately excludes external ratings, other customers' data and any invented signals. A vendor with no history in your workspace has no trust conclusion — which is itself the honest answer for a new counterparty.
The problem with borrowed reputation
Most trust scores in software are borrowed: aggregated reviews, global ratings, risk databases built from other companies' experiences. Borrowed reputation fails for agent vendors in both directions. A globally reputable vendor can still be the one whose pricing page reliably steers your agent into premium tiers. An obscure specialist vendor can serve your workflow flawlessly for a year while carrying no public reputation at all. What you actually need to know is narrower and more valuable: how has this vendor behaved with your agents, under your mandate, on your tasks? That question is answerable only from your own ledger, and MandateGuard's trust view refuses to answer it from anything else.
What the score is computed from
Every authorize decision in your workspace is a signed receipt: vendor, amount, decision, reason. The trust view aggregates those receipts per vendor — counts of allowed and denied decisions and total allowed value — and presents the score beside those raw counts. Showing the workings is a design commitment, not decoration: a score resting on two decisions should be read differently from one resting on two hundred, and a denial caused by your daily cap says something different from a denial caused by an over-cap upsell attempt. Because the underlying ledger is hash-chained, the aggregates are computed over a history you can verify, not a log you have to take on faith.
How to read the patterns
Healthy vendor history is boring: mostly allows, amounts in a stable band, denials that correspond to your own thresholds and were later approved. Three patterns deserve attention. A rising share of over-cap denials suggests the vendor's offers drift upward — packaging changes, default tiers, pre-ticked upgrades. A cluster of duplicate escalations suggests the vendor's checkout flow invites re-purchase, or your workflows do. Denials for amounts that cluster just under your approval threshold can indicate threshold-aware pricing worth a conversation. None of these readings requires the score to be a single magic number; the counts tell the story, and the score simply sorts the table so the stories surface in order.
Trust as an allowlist maintenance tool
The allowlist decides who may be paid; trust tells you how that permission is working out. At a monthly review, sort vendors by activity and ask three questions. Is every listed vendor still needed by a job? Has any vendor's behaviour pattern changed since last review? Is any vendor carrying standing permission with almost no history — permission granted in hope rather than experience? Removing a dormant vendor costs nothing and shrinks the agent's reachable surface. Tightening approvals for a drifting vendor is a proportionate response that keeps the job running. Trust data turns these from vibes into an agenda with numbers attached.
Limits of first-party trust
Own-ledger trust has honest limits. It cannot warn you about a vendor you have never used — for that, ordinary procurement diligence applies, and the allowlist's default-deny posture is your protection while history accumulates. It also cannot see behaviour outside your workspace's transactions: a vendor's treatment of other customers is invisible by design, because pooling it would recreate the borrowed-reputation problem and mix other companies' data into yours. Finally, trust summarises decisions, not outcomes: a vendor can be perfectly allowed and still deliver poor goods. Pair the trust view with your own delivery checks, and let each system answer the question it can actually see.
Where trust sits in the paid depth
Vendor trust is part of Pro and Team, alongside duplicate detection, the kill-switch, evidence export and higher agent and policy limits. It is computed from the same receipt history the free plan already records, so upgrading does not start your history from zero — the ledger your free agent built becomes the basis of the trust table on day one. For teams running several agents, trust aggregates at workspace level, which is the right altitude for vendor decisions: the question is how the vendor behaves with your organisation, and the receipts of every agent you run are the evidence.
Five actions to take now
- Let history accumulate before judging scores — trust is computed from your ledger only
- Review trust beside counts: allowed, denied and totals are shown for a reason
- Use low-trust patterns to prune the allowlist, not to punish vendors automatically
- Recheck trust after price or vendor changes to reset expectations
- Export trust evidence with the receipt pack when finance asks for vendor reviews
Worked example: reading a trust table at renewal time
After two quarters, a workspace's trust view shows three instructive vendors. Its model API provider: 312 decisions, 308 allowed, 4 denied (all approval-threshold escalations later granted), 4,180 dollars allowed in total — a deep, boring, healthy history. Its data broker: 41 decisions, 30 allowed, 11 denied, including repeats of the same over-cap amount — a vendor whose offers keep pushing the mandate. Its cloud provider: 96 decisions, all allowed, 1,940 dollars total.
At renewal time the team does not debate impressions. The broker's pattern justifies renegotiating the purchase shape, the two healthy vendors renew without ceremony, and a fourth vendor with two lifetime decisions is treated as what it is: unknown, and kept on a short approval leash until its own history says otherwise.
Frequently asked questions
Where does the data come from?
Only your workspace's receipts. If your agents have not transacted with a vendor, MandateGuard has nothing to score and says so.
Can a vendor improve its standing?
Yes — by transacting cleanly. Allowed decisions accumulate; the counts beside the score show the basis so you can judge the improvement yourself.
Should low trust auto-deny a vendor?
Trust informs the allowlist and approval decisions; it does not silently replace them. Remove a vendor from the allowlist if you no longer want it payable.
Is trust shared across workspaces or customers?
No. Your ledger is yours. The fleet-wide claim of no invented data means no pooling or borrowed reputation.
Which plans include vendor trust?
Pro and Team. The free plan keeps the full receipt history from which trust is computed if you upgrade later.
Keep reading
- How does a vendor allowlist work for AI agent spending?
- How does duplicate-commitment detection stop agents paying twice?
- What should a monthly AI agent spend review include?
Put the policy in force: spend envelopes · intent registry · approval inbox · how the proof works · all MandateGuard guides