How scoring works

Agent Payments Arena asks two separate questions: how ready is this company for an AI-initiated payment, and how much of that payment flow has actually been proven? This page explains how to read both answers.

The short version

  • Readiness score: How easy and safe the public payment flow appears to be for an AI agent.
  • Evidence level: The strongest result Lightsage has actually verified, from a public claim to a completed payment.
  • Not detected:The scan did not find public evidence for that payment system, so it does not lower the score.

What the score means

Overall readiness is 70% agent access and 30% payment readiness.

  • Payment readiness asks whether an agent can find the payment flow, understand the terms, complete the required steps safely, and receive the expected result.
  • Agent access asks whether public instructions, documentation, and machine-readable interfaces are reachable and understandable.

If a dimension has not been evaluated, it contributes zero until evidence is available. A high readiness score still does not prove that money moved.

What we check

  • x402 and MPP: Can an agent find the paid endpoint, understand an unpaid payment request, pay safely, and receive the promised result and receipt?
  • UCP and ACP: Can an agent discover products, use checkout, submit payment details, and recover or track an order?
  • AP2: Can the merchant verify exactly what the user authorized the agent to buy?

What each result means

  • Passed: The available evidence satisfied the check.
  • Needs fix: The check failed or required information was missing.
  • Needs verification: Some evidence exists, but a static scan cannot prove the complete behavior.
  • Not detected: The scan did not find public evidence for that payment system or check, so it is not scored.

Evidence levels

Every company report shows the highest level supported by its current evidence:

  1. Not evaluated: No conclusion has been reached.
  2. Support claimed: The company says it supports the payment system.
  3. Payment signal detected: Public metadata or documentation suggests support.
  4. Unpaid challenge validated: The endpoint returned a valid payment request, but no funds moved.
  5. Sandbox verified: A complete payment worked using test funds.
  6. Live verified: A limited production payment completed successfully.
  7. Multi-payer verified: More than one compatible payer completed the flow.
  8. Certified: Lightsage reviewed reproducible, current evidence.

How we decide what applies

The Arena looks for a matching payment surface, public metadata, or an explicit support claim before treating a payment system as applicable.

For example, a merchant that does not use AP2 should not lose points for missing AP2 metadata. But if a company claims x402 support and its paid endpoint does not return a usable x402 challenge, that is a failed check.

Freshness and corrections

Each report shows when it was last scanned. A successful rescan replaces the published result after the new evidence is saved. Companies can report an incorrect result from their company page.

Safety

Public scans never authorize payments. Sandbox tests use test funds and fixed budgets. Live tests require approved endpoints, explicit authorization, separate spending limits, and auditable receipts.
Back to Agent Payments Arena