Verification layer for what AgentCore Payments actually bought

[stillmarcus24]

AgentCore Payments solves the hard part of the pay side well — managed wallets, spending policy, an agent that never touches private keys. Nothing in the payments module answers what happens after settlement clears: whether the thing the agent just paid for was actually what it claimed to be. We've been running a live answer to that separately, on Base mainnet, and it looks like a natural pairing with what AgentCore already manages.

The mechanism: a claim gets written into a signed, hash-chained receipt before we know how it resolves. Resolution checks it against a source neither side controls — a merged GitHub PR, a settled Kalshi market, a raw HTTP probe — cited by name in the receipt. Real capital backs every verdict: a $10 USDC bond, $1 slashable per verdict a counterparty proves wrong. A dispute is POST /dispute with the receipt hash and counter-evidence, resolved by an independent re-check against the same source — not our call to make. Bond balance and slash history are open at /notary/bond.

Concretely: an AgentCore-managed agent settles an x402 payment through your PaymentSession, then optionally routes the returned resource through a verification check before the policy-controlled spend is considered "worth it." That's a thin addition on top of what's already built — the spending-limit and audit-trail work AgentCore does for payment already assumes someone eventually asks "did this spend get us something real," and right now nothing answers that.

Real settlement, not simulated: $0.10 on Base mainnet, tx 0x8a9f.... Contract 0xA3a05818d4051BFa759Fb7D936b57C072e4E0Caf, public.

Happy to sketch a concrete sample against the SDK if a post-payment verification step is useful for the samples repos.