The assurance loop, in detail.
Four artifacts, produced around a single agent transaction. Together they answer the question no payment protocol answers: did the buyer get what it paid for, and if not, what happens next.
- 01The four artifacts
- 02Verdict semantics
- 03How it attaches
- 04Evidence grading
- 05What we never see
Preflight and postflight assurance.
One record, inspectable by the buyer, the seller and a regulator, assembled from four artifacts at four points in the transaction.
Preflight score
Qualifies the service against the offer before any value moves, using the live census rather than provider self-description.
- Provider identity
- Price vs market
- Observed latency
- Risk & SLA fit
- Wallet policy fit
Signed record
Binds everything that justified the payment into a single signed artifact, so the decision can be reconstructed later.
- Inferred intent
- Offer & terms
- Spend cap
- Delegation chain
- Payment authorization
Execution telemetry
Captures what actually happened, which is the difference between a promise and a delivery.
- Response metadata
- Realized cost
- Latency & retries
- Errors
- Delivery evidence
Postflight verdict
Confirms outcome quality against the offer, then acts on the payment automatically.
- Outcome quality
- Offer conformance
- Release or hold
- Retry or reroute
- Refund or dispute
A verdict is an input, not a report.
Four closed outcomes, each mapping to an action an agent or a platform can take without a human in the path.
Exception handling stops being a support process and becomes infrastructure. The financial value is lower support cost, fewer manual disputes, faster resolution, and fewer abandoned workflows, and, upstream of all of it, more autonomy an institution is willing to grant.
It attaches to the rail you already use.
We take no position on which settlement standard wins. Assurance sits above all of them.
VERIFY: the preflight endpoint.
One call, before the buy. An agent submits the service it intends to pay for; VERIFY returns a machine-readable verdict with the evidence behind it. Paid per call in USDC over x402: no account, no contract, no onboarding.
Settlement agnostic.
Assurance is orthogonal to how the money moves, so the method transfers to whichever standard a platform commits to. Live on x402 today; MCP, ACP, UCP, AP2 and TAP are next, not done.
Sits beside the rail.
We are not in the settlement path and cannot stop a payment on our own. We produce the evidence and the verdict; the platform decides what to enforce.
Human-in-the-loop by design.
Hard daily capacity, a kill switch, and a full transaction ledger. Autonomous spending controls are a core product requirement, not an afterthought.
A number you can't defend is a number you can't act on.
Every figure carries its provenance. You always know how much weight a value can bear before you put a decision on it.
Taken directly from the source of record: a signed response, a settled transaction, a provider invoice. Defensible to an auditor.
Cross-checked between two independent sources that agree. Defensible to a CFO.
Modeled from observed behavior where no authoritative figure exists. Directionally sound, labeled as such.
Newly observed and not yet corroborated. A lead to investigate, never a basis for enforcement.
Full visibility. Zero content.
Assurance needs to know whether a service performed, not what was said inside it. Those are separable, and we deliberately only take the first.
We say plainly what this can and cannot see. Assurance based on transaction shape cannot judge whether an answer was correct, only whether the service delivered what the offer described. That boundary is exactly why a CISO and a CFO can both sign off.
See it on a real resource.
One call against a live endpoint tells you more than a deck. Paste a resource into VERIFY and read the verdict. For category boundaries, compare agentic service assurance, payment monitoring, LLM observability, and vendor logs.