0xc54ad46d...0ba93d

An AI agent's decision to pay or refuse a knowledge source is controlled by a **Spend Policy** with nine key parameters: (1) **max price per citation** — the ceiling price the agent will pay; (2) **min relevance score** — a threshold below which sources are rejected regardless of price; (3) **bonded-source requirement** — whether the source must post collateral or reputation; (4) **session cap** — total budget limit per research session; (5) **dedup window** — time period to avoid paying twice for the same content; (6) **receipt visibility** — whether payments must be on-chain and auditable; (7) **creator mandate** — whether the source creator has published spend controls the agent must honor; (8) **x402 protocol support** — whether the source accepts nanopayment settlement; and (9) **policy enforcement mode** — whether spend controls are advisory or hard-blocked. These parameters are typically expressed as a JSON policy document, but can also be generated from plain-English instructions using natural language policy builders. The agent evaluates each candidate source against this policy: if price exceeds max, relevance falls below threshold, or mandatory controls are violated, the source is refused; otherwise, the agent pays via on-chain USDC settlement on Arc with Circle Gateway x402, generating a verifiable receipt tied to the citation.