posted 2026-07-26 · b2a2c.vin/blog/what-should-an-ai-agents-receipt-look-like-after-it-buys-something-for-me

What should an AI agent's receipt look like after it buys something for me?

A receipt you read line by line, not a log you take on faith: evidence consulted, actions taken, price settled under the ceiling, and a dispute/refund URL.

An AI agent's receipt should be something you read line by line — not a log you take on faith. Field for field, it carries four things: the evidence the agent consulted, each fact rendered as value · source · timestamp · verification and byte-identical to what the agent actually fetched; the actions it took, each stamped with who acted and under what authority; the price it settled under the ceiling, showing the amount and the Mandate limit it sat beneath; and a dispute/refund URL that resolves to a real remedy. The test is simple: if the receipt can't show its work — the verified Listings it shopped, the ceiling it held, the gates it settled — then delegation has quietly become dispossession, and the first disputed deal is the last. Auditable by construction means every line on the receipt traces back to a fact at its own address.

That is the difference between a receipt and a log. A log is the agent's narration of what it says it did; you either believe it or you don't. A receipt is a set of claims each of which you can re-open at its source and check independently. The whole point of handing a car moment to an agent is that you stop having to sit at the desk — but if the hand-back is unverifiable, you have not left the desk, you have just moved it inside your own tool. So the receipt's job is to be checkable without the agent's cooperation.

Field one: the evidence consulted

Every fact the agent read appears as an evidence object — value, source, timestamp, and verification status — and it is byte-identical across representations. The value the agent reasoned over is the value you can open at the record's address; the source names where it came from; the timestamp says when; the verification says how it was confirmed. This matters because it closes the gap where an agent could act on a fact it never actually held. If the agent shopped Listings, each Listing on the receipt carries its venue, source, and timestamp, and it is one of the 2.1M+ verified Listings the record resolves — not a screenshot, not a summary, not a number with no provenance. You are not reading the agent's memory of the evidence. You are reading the evidence.

Field two: the actions taken, with authority

Each action carries who acted and under what authority, because in this motion the actor changes from line to line. A read line carries the agent under its Mandate's read scope. A settle line carries the agent, acting for you, under the Mandate ceiling. A handoff line carries you — the principal — under your own signature, at the one step the law or the money reserved for a human. A receipt that flattens all of that into "the agent did X" hides the single most important fact about a delegated act: whether the machine acted or you did. Keeping authority on every line is what lets you confirm that the agent never signed something only you were allowed to sign.

Field three: the price, settled under the ceiling

The money line shows the amount settled and the Mandate ceiling it sat under, and the relationship between them is enforced, not asserted. The ceiling rides inside the typed OFFER and the x402 settle; a charge above the ceiling doesn't get explained away on the receipt, because it never settles in the first place — the settlement fails at the wire. So the price line is not the agent telling you it stayed in budget; it is the residue of a rail that could not have exceeded the budget. Where the deal deepened a fact for a real cost — a per-report OFFER the agent settled to confirm something the deal turned on — that appears as its own line with its own price and its own source, so a metered read is visible as a deliberate step rather than a silent toll.

Field four: the dispute/refund URL

The last field is the one that makes the first three honest: a dispute/refund URL that resolves to an actual remedy. A receipt without a dispute path is a claim of finality the principal never agreed to. With one, every other line becomes a thing you can contest against its own source — the value against its provenance, the action against its authority, the price against its ceiling. The stakes are plain and worth naming: "trust the agent" is the answer that ends the account. A receipt that ends in a real dispute URL is the structural refusal of that answer. The check digit certifies the VIN; it never sells the car — and the receipt certifies the act; it never asks you to take it on faith.

Where the receipt's lines come from

The reason every line can trace to its fact is that the underlying record is already a response built from evidence objects. Each field is a value with a source, a timestamp, and a verification, priced on the page where it is metered — the record as a response, not a marketing surface. That surface, and the keys an agent uses to read and settle against it, live at apis.vin, the transactor the whole motion routes through. And a receipt line with a real cost behind it looks like a vehicle history report: a per-report OFFER an agent settles under the ceiling, its verification and provenance carried on the line, available at vhr.vin. The receipt itself — rendered free for any VIN, keyless, with no account — is the wiring diagram at b2a2c.vin: read, act, return, and a hand-back the principal reads line by line.

The record, at its other addresses

The piece states it; the record posts it. Read yours: