Decentralised News · Evidence reviewed 12 November 2026 · Negotiated terms, authority and timing
What Matters
Two software participants can exchange a quote and payment without proving that two autonomous agents negotiated. Measure the actual terms, the evidence of delegated authority and the events between request and settlement. Milliseconds do not have a universal basis-point price. DN's timing record separates observations and assumptions without claiming a measured agent market or a market-purity score.
DN Evidence Block
Documented: A2A communication, AP2 mandate structures, x402 payment verification and a labelled Stripe/AWS prototype. Original analysis: negotiation classification and event-timing records. Not measured: agent-attributed trading volume, negotiated savings, latency distributions or adoption rankings. “OTC” is a bilateral-commerce analogy; this article does not establish the first such market or a regulated financial venue.
A Quote Is Not Yet a Negotiation
A buyer can request an API price, accept posted terms and pay programmatically. That is useful automation, but it does not establish bargaining. Evidence of negotiation identifies a material term the participants could change, the proposals exchanged and the final accepted version.
Terms extend beyond price. Compute requests can specify hardware, duration and region; datasets have coverage and freshness; procurement includes quantity, delivery and remedies. A lower price for a different service is not a saving on the same service. If a bulk discount changes quantity, or a lower price accepts later delivery, preserve that change in the comparison.
The decision is whether the complete accepted terms satisfy the principal's requirements. Keep the request and outcome together rather than reporting only the favorable movement in one price.
The Protocol Stack Has Distinct Jobs
| Layer | Documented function | Additional evidence |
|---|---|---|
| A2A | Agent Cards, messages, tasks and artifacts support implementation-to-implementation interaction. [1] | That identified participants negotiated the terms and delivered the task. |
| AP2 | Checkout and Payment Mandates describe approval and payment authorization structures. [2] | That relevant signatures, constraints and policy were checked for this transaction. |
| x402 | HTTP 402 communicates payment requirements; an optional facilitator verifies and settles payment payloads. [3] [4] | Agency, negotiated terms, service delivery and refund or dispute rights. |
| ERC-8004 | A draft provides registration, feedback and validation interfaces. [5] | Meaning and relevance of history, plus current principal authority. |
A valid payment payload does not prove successful service delivery. Signed capability information does not establish past performance. No single protocol supplies every business rule, permission and operational guarantee.
Read AP2's Current Mandates Precisely
The reviewed AP2 specification distinguishes a Checkout Mandate from a Payment Mandate, and direct human-present operation from autonomous human-not-present operation. In the latter, an open mandate can constrain what an agent later closes. Payment authorization is bound to relevant checkout information. Implementations must verify the applicable signatures, relationships and constraints; the presence of JSON is not a successful verification result. [2]
A negotiation envelope identifies permitted counterparties, maximum spend, quantity or service boundary, validity and escalation. If final terms fall outside it, new authorization may be required. The specification does not determine the enforceability of a particular contract or make an agent a legal person.
The earlier article said an agent could not technically or legally enlarge its authority. The useful implementation requirement is to reject requests outside the verified permission. Software can produce an unauthorized request; enforcement and legal consequences are separate questions.
What the Stripe Demonstration Establishes
Stripe and AWS's Sessions 2026 presentation showed a buyer-agent flow through different seller configurations. Its transcript explicitly calls this a prototype rather than production. It reports one successful example taking roughly three seconds and costing 12 cents, and discusses seller-side dynamic quotes and quantity negotiation. These are attributed demonstration details, not a DN latency sample, prevailing market price or aggregate autonomous volume. [6]
The practical lesson is the dependency across discovery, usable terms, policy and payment. A fast payment step cannot rescue an incompatible service, missing authority or an expired offer.
Separate the Clocks Before Pricing Delay
| Event | Record | Interpretation |
|---|---|---|
| Request issued | UTC timestamp, request version and terms. | Starts the declared quote-response clock. |
| Quote received | Receipt time, offer identity, terms and expiry rule. | Defines response delay and later offer age. |
| Authority checked | Mandate/policy reference and check time. | Records what was reviewed; does not prove authorization by itself. |
| Offer accepted | Accepted version and time. | Tests acceptance against the actual validity rule. |
| Payment confirmed | Rail-specific event and finality assumptions. | Ends the chosen payment clock. |
| Outcome accepted | Delivery and acceptance evidence. | Ends the business-task clock. |
Clock synchronization, precision and event location matter. A server's send time, buyer's receipt time and block timestamp are not interchangeable. Keep the clock source and uncertainty. Whole-second timestamps do not become millisecond measurements by adding zeros.
Record failed requests and no-quotes alongside successful responses. A median restricted to returned quotes excludes timeouts; publish that denominator and the timeout rule. Distinguish a retried negotiation from a duplicate payment or receipt.
Milliseconds Are Not Automatically Basis Points
For a quote-to-acceptance price comparison, first establish equivalent quantity, asset, quality, fees and settlement terms. For a buy, relative improvement is 10,000 × (reference all-in cost − accepted all-in cost) ÷ reference all-in cost. For a sell, reverse the numerator because higher proceeds are favorable. The reference must be positive, dated and a justified available alternative.
Constructed example: an equivalent purchase costs 100 units at the reference and 99.98 units at acceptance. The difference is 0.02 units, or 2 basis points. If 200 milliseconds elapsed, that does not establish that every millisecond was worth 0.01 basis points. Inventory, quote selection, bargaining, fees and unrelated market movement can all change price. Causal latency analysis needs controlled variation or a justified empirical design.
A separate scenario can price waiting under a chosen assumption. If 100,000 units of capital are assigned an opportunity cost of 2 basis points per hour and remain unavailable for 30 seconds, the linear result is 100,000 × 2/10,000 × 30/3,600, approximately 0.1667 units. That is an assumed waiting cost, not an observed debit, yield forecast or proof that negotiation caused a loss. Do not add it to a price loss representing the same effect.
Classify Transactions From Evidence
| Classification | Minimum evidence | Unknown without further evidence |
|---|---|---|
| Machine-compatible payment | Identifiable rail and payment record. | Whether an autonomous agent chose the purchase. |
| Agent-to-service purchase | Buyer-agent decision and service-outcome evidence. | Whether the seller used an agent or negotiated. |
| Bilateral agent transaction | Both roles and the material interaction. | Whether terms changed or humans intervened. |
| Negotiated bilateral transaction | Proposals, accepted terms, authority and outcome references. | Universal autonomy level or regulatory status. |
“Not verified as A2A” does not mean human-only. An agent-compatible endpoint also does not identify its beneficial owner. Define the cohort's unit, window, classification rules, missing-evidence category and review process before publishing proportions. Transaction-count shares and value shares are different statistics.
The former market-purity calculator multiplied user-entered percentages and labelled results emerging or high-purity market activity. Arithmetic could not establish those conclusions. The replacement record contains no adoption or market-size verdict. It preserves unknowns instead of turning assumed shares into attributed autonomous volume.
Protect Negotiating Limits and Preserve Accountability
A maximum budget is an enforcement boundary, not necessarily information a seller needs. Preserve relevant decisions and authorization evidence without publishing confidential reservation values, personal identifiers or credentials. Record who may inspect the underlying material and how disputed findings can be corrected.
Related-party trades, promotional credits and repeated self-funded transactions can distort a volume sample even when payments occur. Keep those categories visible. A transaction graph alone is not an ownership map. Payment confirmation is also separate from the remedy if service is late, unusable or never delivered.
Credit and other regulated financial transactions require evidence beyond the protocol checks described here. Reputation can inform a counterparty review, but neither a payment nor a favorable history establishes authority for the next obligation.
Use the DN Negotiation Timing and Evidence Record
The manual template below records terms, participant evidence, event clocks, authority, comparison baseline, outcomes and missing cases. It is not a stopwatch, automated classifier or market-data feed. Use non-sensitive reference labels and state the units and assumptions in notes. Completing the schema produces no verified result.
For an empirical study, collect a consented sample of comparable requests, retain no-quotes and expired offers, and predeclare the baseline and timing endpoints. Separate service quality from price, and classify human intervention from logs rather than guessing from a wallet. Only then can the cohort support a bounded latency or negotiated-outcome statistic.
Continue the Research
Best Wallets for AI Agents · x402 Cost & Reliability Benchmark · Agent Checkout Reliability Index · Best Crypto Platforms for AI Agents. Related decision pages may contain separately disclosed commercial relationships.
Frequently Asked Questions
Does x402 volume prove negotiated agent commerce?
No. The rail does not establish both roles, bargaining, authority or delivery.
Can latency be converted directly to basis points?
No universal conversion exists. Record elapsed time and comparable price outcomes separately; causal attribution requires more evidence.
Is the Stripe example a production benchmark?
No. The cited transcript explicitly identifies a prototype demonstration.
Methodology and Change Log
12 November 2026: updated AP2 terminology and prototype scope; replaced uncalibrated market-purity outputs with a versioned timing/evidence record; separated delivery from payment and elapsed time from causal price effects. Negotiation-envelope, service-term, classification and accountability concepts are retained. No live negotiation, financial transaction or independent latency test was performed.
Corrections should identify the protocol revision, event definition, source record and affected claim. Do not send confidential negotiating limits or credentials. Revisit when consented and reproducible observations support an empirical cohort.
Machine-readable Negotiation Timing and Evidence Record
Schema urn:dn:negotiation-timing-evidence:v1, version 1.0.0. Copy the blank JSON below and retain non-sensitive references for the defined review. All unknowns remain explicit. Change basis to self_reported_review and supply a real calendar review date and subject when completing it. The schema checks structure only; it does not fetch or authenticate a reference. No API or live dataset is implied.
| Area key | Meaning |
|---|---|
participantRoles | Buyer and seller agency evidence; unknown roles remain unknown. |
termsAndVersions | Request, quote and accepted terms; quantities, fees, quality and validity. |
eventClocks | Request, quote receipt, authority check, acceptance, payment and delivery times; timezone, clock and uncertainty. |
authority | Principal, mandate/policy version, constraints and verification evidence. |
baseline | Comparable alternative, timestamp, units, availability and confounders. |
outcome | Payment and delivery separately; accepted, partial, disputed or pending. |
cohortAndMissingness | Unit, window, eligibility, timeouts, retries, no-quotes and missing evidence. |
privacyAndRecourse | Permitted access, sensitive-data handling and correction/dispute path. |
Each area has status (unknown, reference_recorded or issue_recorded), reference and notes. A recorded reference or issue requires a non-empty reference. Unknown means unresolved; it is not an observed zero or a failure finding. Preserve private source documents in their authorized system rather than embedding account numbers or credentials.
Copy blank record JSON
{
"recordType": "negotiation-timing-evidence",
"schemaVersion": "1.0.0",
"basis": "blank_template",
"reviewedDate": null,
"subject": null,
"areas": {
"participantRoles": {
"status": "unknown",
"reference": null,
"notes": null
},
"termsAndVersions": {
"status": "unknown",
"reference": null,
"notes": null
},
"eventClocks": {
"status": "unknown",
"reference": null,
"notes": null
},
"authority": {
"status": "unknown",
"reference": null,
"notes": null
},
"baseline": {
"status": "unknown",
"reference": null,
"notes": null
},
"outcome": {
"status": "unknown",
"reference": null,
"notes": null
},
"cohortAndMissingness": {
"status": "unknown",
"reference": null,
"notes": null
},
"privacyAndRecourse": {
"status": "unknown",
"reference": null,
"notes": null
}
},
"boundary": "This record does not independently verify evidence, measure performance or authorize an action."
}
Documented JSON Schema
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"$id": "urn:dn:negotiation-timing-evidence:v1",
"title": "Negotiation Timing and Evidence Record",
"description": "An entered evidence record. Schema validation checks structure, not authenticity, legal permission or physical or financial outcomes.",
"type": "object",
"additionalProperties": false,
"required": [
"recordType",
"schemaVersion",
"basis",
"reviewedDate",
"subject",
"areas",
"boundary"
],
"properties": {
"recordType": {
"const": "negotiation-timing-evidence"
},
"schemaVersion": {
"const": "1.0.0"
},
"basis": {
"enum": [
"blank_template",
"self_reported_review"
]
},
"reviewedDate": {
"type": [
"string",
"null"
],
"pattern": "^\\d{4}-\\d{2}-\\d{2}$",
"format": "date"
},
"subject": {
"type": [
"string",
"null"
],
"maxLength": 240
},
"areas": {
"type": "object",
"additionalProperties": false,
"required": [
"participantRoles",
"termsAndVersions",
"eventClocks",
"authority",
"baseline",
"outcome",
"cohortAndMissingness",
"privacyAndRecourse"
],
"properties": {
"participantRoles": {
"type": "object",
"additionalProperties": false,
"required": [
"status",
"reference",
"notes"
],
"properties": {
"status": {
"enum": [
"unknown",
"reference_recorded",
"issue_recorded"
]
},
"reference": {
"type": [
"string",
"null"
],
"maxLength": 1000
},
"notes": {
"type": [
"string",
"null"
],
"maxLength": 2000
}
},
"allOf": [
{
"if": {
"properties": {
"status": {
"enum": [
"reference_recorded",
"issue_recorded"
]
}
}
},
"then": {
"properties": {
"reference": {
"type": "string",
"minLength": 1
}
}
}
}
]
},
"termsAndVersions": {
"type": "object",
"additionalProperties": false,
"required": [
"status",
"reference",
"notes"
],
"properties": {
"status": {
"enum": [
"unknown",
"reference_recorded",
"issue_recorded"
]
},
"reference": {
"type": [
"string",
"null"
],
"maxLength": 1000
},
"notes": {
"type": [
"string",
"null"
],
"maxLength": 2000
}
},
"allOf": [
{
"if": {
"properties": {
"status": {
"enum": [
"reference_recorded",
"issue_recorded"
]
}
}
},
"then": {
"properties": {
"reference": {
"type": "string",
"minLength": 1
}
}
}
}
]
},
"eventClocks": {
"type": "object",
"additionalProperties": false,
"required": [
"status",
"reference",
"notes"
],
"properties": {
"status": {
"enum": [
"unknown",
"reference_recorded",
"issue_recorded"
]
},
"reference": {
"type": [
"string",
"null"
],
"maxLength": 1000
},
"notes": {
"type": [
"string",
"null"
],
"maxLength": 2000
}
},
"allOf": [
{
"if": {
"properties": {
"status": {
"enum": [
"reference_recorded",
"issue_recorded"
]
}
}
},
"then": {
"properties": {
"reference": {
"type": "string",
"minLength": 1
}
}
}
}
]
},
"authority": {
"type": "object",
"additionalProperties": false,
"required": [
"status",
"reference",
"notes"
],
"properties": {
"status": {
"enum": [
"unknown",
"reference_recorded",
"issue_recorded"
]
},
"reference": {
"type": [
"string",
"null"
],
"maxLength": 1000
},
"notes": {
"type": [
"string",
"null"
],
"maxLength": 2000
}
},
"allOf": [
{
"if": {
"properties": {
"status": {
"enum": [
"reference_recorded",
"issue_recorded"
]
}
}
},
"then": {
"properties": {
"reference": {
"type": "string",
"minLength": 1
}
}
}
}
]
},
"baseline": {
"type": "object",
"additionalProperties": false,
"required": [
"status",
"reference",
"notes"
],
"properties": {
"status": {
"enum": [
"unknown",
"reference_recorded",
"issue_recorded"
]
},
"reference": {
"type": [
"string",
"null"
],
"maxLength": 1000
},
"notes": {
"type": [
"string",
"null"
],
"maxLength": 2000
}
},
"allOf": [
{
"if": {
"properties": {
"status": {
"enum": [
"reference_recorded",
"issue_recorded"
]
}
}
},
"then": {
"properties": {
"reference": {
"type": "string",
"minLength": 1
}
}
}
}
]
},
"outcome": {
"type": "object",
"additionalProperties": false,
"required": [
"status",
"reference",
"notes"
],
"properties": {
"status": {
"enum": [
"unknown",
"reference_recorded",
"issue_recorded"
]
},
"reference": {
"type": [
"string",
"null"
],
"maxLength": 1000
},
"notes": {
"type": [
"string",
"null"
],
"maxLength": 2000
}
},
"allOf": [
{
"if": {
"properties": {
"status": {
"enum": [
"reference_recorded",
"issue_recorded"
]
}
}
},
"then": {
"properties": {
"reference": {
"type": "string",
"minLength": 1
}
}
}
}
]
},
"cohortAndMissingness": {
"type": "object",
"additionalProperties": false,
"required": [
"status",
"reference",
"notes"
],
"properties": {
"status": {
"enum": [
"unknown",
"reference_recorded",
"issue_recorded"
]
},
"reference": {
"type": [
"string",
"null"
],
"maxLength": 1000
},
"notes": {
"type": [
"string",
"null"
],
"maxLength": 2000
}
},
"allOf": [
{
"if": {
"properties": {
"status": {
"enum": [
"reference_recorded",
"issue_recorded"
]
}
}
},
"then": {
"properties": {
"reference": {
"type": "string",
"minLength": 1
}
}
}
}
]
},
"privacyAndRecourse": {
"type": "object",
"additionalProperties": false,
"required": [
"status",
"reference",
"notes"
],
"properties": {
"status": {
"enum": [
"unknown",
"reference_recorded",
"issue_recorded"
]
},
"reference": {
"type": [
"string",
"null"
],
"maxLength": 1000
},
"notes": {
"type": [
"string",
"null"
],
"maxLength": 2000
}
},
"allOf": [
{
"if": {
"properties": {
"status": {
"enum": [
"reference_recorded",
"issue_recorded"
]
}
}
},
"then": {
"properties": {
"reference": {
"type": "string",
"minLength": 1
}
}
}
}
]
}
}
},
"boundary": {
"const": "This record does not independently verify evidence, measure performance or authorize an action."
}
},
"allOf": [
{
"if": {
"properties": {
"basis": {
"const": "self_reported_review"
}
}
},
"then": {
"properties": {
"reviewedDate": {
"type": "string"
},
"subject": {
"type": "string",
"minLength": 1
}
}
}
}
]
}
Related reading:
Best Agentic Checkout Systems 2027: ACP vs UCP vs Visa vs Mastercard
Fastest Crypto Trading APIs 2027: Latency, Rejections & Reliability Ranked
Best Crypto Platforms for AI Agents 2027 | Agentic Finance Rankings
Agentic Wallet Security Index 2027: The Safest Wallets for AI Agents
Best Wallets for AI Agents 2027: Security, Payments & Autonomous Finance




