GUIDE / AGENT TRANSACTIONS
What counts as an agent transaction?
Money moved. But did an agent buy something, and did it work? Each claim needs its own evidence.
Four things to check
- Was there a payment? Check who received it, the currency, amount and payment status.
- Was it an authorized service purchase? Match the payment to the buyer’s request, agreed price and terms. A transfer alone is not enough.
- Was the agreed output delivered? Check the delivered file or answer against what was promised. A server response alone is not proof.
- Did the result satisfy the task? Test the result against the buyer’s requirements. A readable file can still contain a wrong answer.
A concrete example
An agent pays for a weather report. The payment shows money moved. A linked receipt shows what was bought. A file check shows which report arrived. Only checking its city, date and source tells us whether it answered the request.
Why transaction volume is not agent revenue
Transfers include fees, trading and moving funds between wallets. They are not all purchases. Counting every transfer as agent revenue would overstate the market.
What InferenceView can show today
We show recorded Tempo transactions, recognized amounts and exchanges, and evidence checks. We do not yet verify agent identity, delivery or task quality. Explore the dataset.
What would improve confidence?
Buyer requests, seller receipts and independent result checks would help. A long payment history alone does not prove that customers received useful results.
Go deeper: x402 payments, AP2 authorization, and our scoring method.
