DN research framework · 18 November 2026 · Forward-looking 2027 monitoring
What matters
The best agentic-finance scoreboard counts completed economic outcomes, not promotional launches, request logs or repeated simulated transfers. Our ten indicators below define observable evidence, the minimum denominator and the counterevidence that would change the thesis. This is a monitoring framework, not a claim that these ten series have already been measured.
Ten signals and falsifiers
| Indicator | Evidence to collect quarterly | What would change our mind |
|---|---|---|
| 1. Settled value attributable to agents | Completed payments with identified payer/agent initiation and settlement proof | Growth is driven by bot self-payments or test traffic |
| 2. Real merchants accepting agent checkout | Unique active merchants with at least one confirmed fulfilled order | Payments accepted but orders cancelled or merchants inactive |
| 3. Approval and dispute outcomes | Ratio of successful authorized purchases to failed/disputed attempts | High decline, cancellation or fraud cases persist |
| 4. Cost per completed transaction | All-in fees, retries, model inference and human review per settled order | AI handling costs exceed alternatives after retries |
| 5. Real adoption, not demo counts | Production users and recurring completed tasks, with dates | Adoption is driven by repeated prompts or sandbox-only accounts |
| 6. Transaction permission granularity | Share of deployments enforcing scope, spend cap, expiry and approved counterparties | Agents require blanket signing authority |
| 7. Cross-border payout success | Destination-received payments and dispute rates per corridor | High initiated count hides stuck funds and failed payouts |
| 8. AI error and exception burden | Human correction hours and loss events per 1,000 executions | Increasing volume drives faster exception growth |
| 9. Independent interoperability | Transactions successfully handled across providers and rails | Only proprietary closed-loop demos work |
| 10. Auditability and reversibility | Presence of signed authorization, receipts, recovery and reproducible logs | Settled claims cannot be independently reconciled |
Quarterly DN Scoreboard: evidence status
Baseline status (18 November 2026): Pending independent collection. We will not fill an empty series with inferred totals. Each quarter should record a metric owner, source document or chain query, population, date range, definition, revision history and limitations. Until those fields exist, its status is unscored.
DN Data Integrity Instrument
What does not count
- An HTTP payment-required or access challenge is not a transfer of value.
- An agent-created wallet is not necessarily an economically active user.
- A claimed transaction may be a testnet, internal transfer, self-payment or reversed order.
- Number of integrations does not prove that buyers and merchants can settle an end-to-end purchase.
- A product press release does not prove the feature is available in a given region.
Method and data licence
A future numerical series should release a CSV or JSON record with metric ID, measurement period, value, units, primary sources, exclusion rule and verification date. Any vendor-submitted data should be identified as vendor-reported and independently verified where feasible. For comparison, see DN’s existing agent payment research, and treat the latest payment-wallet launches as dated availability, not a global usage denominator.
Sources and caveats
- OpenAI ChatGPT Wallet launch, 13 November 2026: capability and initial eligibility, not proof of transaction volume.
- x402 official developer documentation: payment protocol mechanics, not transaction-value estimates.
- NIST AI Risk Management Framework: evidence, governance and risk handling context.
Disclosure: No dataset, adoption percentage, settlement amount or 2027 forecast is represented as independently measured in this article. Sponsorship and affiliate relationships, if any, do not influence the proposed indicators. Educational research, not investment advice.
Frequently asked questions
Is x402 traffic the same as settled payments?
No. Access requests, authorization challenges and completed paid settlements are different units. The metric must state its transaction definition and source.
Which single metric matters most?
Completed settled economic value plus the cost and exception rate required to produce it. No metric alone proves sustainable adoption.
Does this page forecast a 2027 market size?
No. It defines observable signals and falsifiers. No unverified market size is presented.




