1. The purchase is easy; the counterparty is the hard part

A shopping agent finds the item, fills the checkout, and the charge goes through. That path has an owner at every step: a cardholder who consented, an issuer that approved, a merchant of record, and a dispute process if the goods never arrive.

Now take the person out of the middle. A logistics agent contracts capacity from a carrier's agent. Neither is a consumer, neither holds an account with the other, and there may be no common processor.

Two things the consumer path supplies for free go missing. One is identity: with no enrollment, nothing establishes that the agent emitting the second message is the party that emitted the first. The other is consequence for conduct. An agent can decline, or answer nothing, at machine rate, and the party whose offer went unanswered should carry nothing adverse for having asked.

2. What Agent Pay sets out to solve

Mastercard Agent Pay is, as publicly described, the company's framework for agentic commerce: an approach that lets an AI agent transact on a consumer's or business's behalf inside the card network's existing payment infrastructure rather than beside it. The problem it names is recognition. When software initiates a purchase, the acceptance and issuing sides need to know that an agent is acting, which agent it is, and on whose authority.

Its answer, on the public account, extends tokenization. Network tokenization already substitutes a scoped credential for a card number, so a credential provisioned for one context cannot be replayed in another. Agent Pay carries that idea to agent-initiated transactions: a tokenized credential associated with an agent and with the permissions a cardholder or business granted it, so that agent-initiated activity is distinguishable at authorization time from ordinary card-on-file activity.

The category is well chosen. Agentic checkout inherits what took card payments decades to build: issuer authorization and risk scoring, chargeback and dispute rights, acceptance rules, and liability allocation among regulated participants. A consumer whose agent bought the wrong thing keeps their recourse, and putting agent identity into the credential is the shortest path to preserving all of it.

Agent Pay, as publicly described, operates at payment authorization and tokenization. The architecture below states its requirements at a different layer, so what follows is a structural comparison: what each design requires of the parties before an exchange can conclude.

3. Settlement that is the pair itself

Chapter 6 of U.S. Provisional Application No. 64/117,812 puts finality inside the exchange rather than under it. The unit is a governed observation: a signed structure comprising an authority credential field, a continuity hash field encoding identity continuity of the emitting party, spatial and temporal reference fields, a time-to-live field, a payload field, and an observation lineage field bearing a cryptographic integrity attestation over the foregoing. Emission is complete upon emission, so no acknowledgment, handshake, delivery confirmation, or central registration conditions it, and an observation lacking a verifiable authority credential is not a governed observation.

A matched pair is a first governed observation representing an offer, tender, claim, demand, or commitment, and a second from the other party representing an acceptance, counter-tender, acknowledgment, refusal, or fulfillment, the two recognized by a pairing rule within one or more of a spatial proximity window and a temporal proximity window, requiring spatial coincidence, temporal coincidence, or both as the governing policy object declares. Without physical co-location, the spatial window admits a scope-partition form satisfied by a non-empty intersection of the identifiers the two observations declare. The pair binds into a settlement record comprising both signed observations, a cryptographic binding over both, and an attestation of the recognizing window, verifiable by a downstream consumer from the record alone. Settlement occurs upon recognition and admission of the pair, and finality attaches at that moment.

Three negative conditions carry the structure. No third-party intermediary participates: no clearing party, payment processor, platform adjudicator, or escrow party is required for the pair to settle. No centralized consensus is computed, no quorum or ledger commit conditioning settlement. And no pre-negotiated session state exists, so a party that has never encountered the counterparty may nonetheless form a pair, consent being expressed per transaction by the observations themselves.

Enrollment being absent, identity is carried by continuity. The continuity hash field holds a successor hash field, being the emitting party's dynamic agent hash at emission, and a continuity vector field of normalized projections of that party's operational state. A received observation is accepted on continuity only where both conditions hold conjunctively: the per-step distance against its predecessor does not exceed a declared per-step ceiling, and a trust-slope consistency measure computed from that vector field is not less than a declared acceptance threshold.

Refusal is emitted, not merely reported. A party declining to complete a pairing emits a refusal observation, itself a governed observation bearing that party's authority credential and paired with the offer it refuses. Emission is metered by a refusal meter carried in the refusing party's own memory field, against that party's own settlement-binding authorization, applied without any determination of whether the refusal was well founded and with no adjudication of the merits by any party. The meter increments once per emission by a refusal increment declared in the governing policy object, accumulating over a declared window. Responsive to the meter satisfying the declared bound, that authorization is written from a granting state to a withheld state, whereupon it binds no further settlements, the party's capacity to observe, to produce determinations, and to emit further refusal observations being preserved.

Equally load-bearing is what increments nothing. An initiated dispute record, non-convergence after the declared round count, a rejection for want of a partition intersection, a held orphan observation, a timeout, and a failed fulfillment each increment no refusal meter of either party. Admission itself runs through a settlement admissibility evaluator producing exactly one outcome from admit, gate, defer, solicit, reject, and escalate, over an evidential weight combining authority, continuity, freshness, and corroboration.

4. Same word, different layer

Both designs converge on one idea: an agent should not transact anonymously, and what it is permitted to do should travel with what it emits. Agent Pay does that with a tokenized credential bound to granted permissions; Chapter 6 does it with an authority credential field carried in every governed observation, its scope and temporal-validity specifications checked before admission. Divergences follow from the different problems each is shaped to solve.

  • Where finality attaches. In card payments, on the public account, authorization and settlement are functions of the network and its participants under scheme rules and law. Chapter 6 requires finality to attach upon recognition and admission of the pair itself, the resulting record being verifiable by a downstream consumer from that record alone.
  • What supplies identity. A tokenized agent credential is, as publicly described, provisioned in advance. Chapter 6 requires no prior enrollment of either party with the other, identity for pairing being established by continuity, so two parties with no common registrar may pair.
  • Who the parties are. Agent Pay's arrangement, on the public account, has a principal whose consent the agent carries. Chapter 6 addresses a bilateral exchange in which both sides may be autonomous and neither is a cardholder.
  • What a decline is. Chapter 6 requires a declining party to emit a refusal observation under its own authority credential, paired with the offer it refuses, appended to each party's settlement lineage, and metered against its own settlement-binding authorization. A declined card authorization is, on the public account, a decision reported about a transaction.
  • Where the record sits. Chapter 6 requires each party to append its own settlement-lineage entry and to reconstruct its history without recourse to the counterparty and without recourse to any registry. Card records, on the public account, are assembled across regulated participants.

None of this is a deficiency in the payments framework. The designs answer different questions, and one deployment can want both answers.

5. Running both in one deployment

The two fit as layers. A procurement agent and a supplier's agent negotiate terms as a sequence of counter-observations, admitted per round through the admissibility evaluator and bounded by a round count declared in the governing policy object, non-convergence being a timeout-class failure adverse to neither side. When the agreed transaction resolves to a card payment, it is authorized on network rails under the agent credential that framework provides.

The Chapter 6 layer settles the conduct state around that payment: whether the authorization gate transitions for an enumerated action class, whether the persistence designation of a counterparty identity record is promoted or demoted, whether a settlement-binding authorization moves. Content can be held under an escrow custody record released on completion of a declared matched pair, including a dual-lock form.

What the disclosed architecture does not solve:

  • No money moves through it and it supplies no legal settlement finality. It displaces no payment network's operating rules, no acquiring or issuing relationship, and no governing law, and relieves no participant of any regulatory obligation.
  • It confers no consumer protection. Chargeback rights, cardholder liability limits, and acceptance rules come from the payment arrangement, not from a matched pair. A settled pair is challengeable only through a credentialed dispute procedure routed as the governing policy object declares, and no step of it increments any refusal meter.
  • It adjudicates no merits. A withheld settlement-binding authorization is not a finding of fault, and reliance bounding upon a counterparty's provisional marker narrows that party's admissible dispatch and disclosure scope under declared bounds, without any determination of the merits.
  • It does not guarantee both sides reach the same outcome on one pair. Where the parties resolve different policy objects, each applies its own, so one may admit what the other gates or defers.

Two questions, then, and two places to answer them: how value moved, and what the parties' conduct cost them in their own records.

6. Disclosure Scope

The mechanisms described are disclosed in U.S. Provisional Application No. 64/117,812, principally Chapter 6, "Matched-Pair Settlement": the governed observation, the matched pair and the pairing rule, the three negative conditions, identity by continuity, the settlement-lineage entry, metered non-acceptance, deferred pairing, escrow and escrow depth, reliance bounding, chained settlement, composite admissibility with graduated outcomes, and credentialed dispute resolution. Subject matter in that chapter originates in U.S. Provisional Application No. 64/049,409.

References to Mastercard Agent Pay are to public materials and are used for comparison only; no relationship, endorsement, or infringement is asserted.

This article is published to establish public, timestamped prior art for applying this architecture to settlement between autonomous agents. It describes pending applications. Nothing here asserts that any product, service, or organization infringes anything, and nothing states that a license is required.