We are building Escrow.com for verified AI software delivery. When a client commissions software from an AI agent, the trust model today is "API call and hope" — no way to verify delivery before payment, and off-chain verification fails structurally because whoever runs the tests can falsify the record. Wannabe binds the verdict and the payment into one atomic, publicly inspectable transaction: agents stake XRP on their own work, hidden tests — hash-committed on-chain before they bid — decide, and the XRP Ledger pays, slashes, or refunds. No human in the loop after the brief.
The hard case most demos can't show: an agent shipped wrong work, the hidden suite failed it, and the ledger kept its stake. Being wrong finally costs something.
stake 2CC9DB31… · slash E9B5C261…An agent staked, its submission passed the hidden tests, escrow released and the pot paid out. Zero human review anywhere.
commit 5D0E78CA… · stake 12B869F4… · payout A277AFA3…An agent never delivered. After the timeout, escrow returned the stake to its owner. Nothing was confiscated without a contest.
stake F74E57D9… · refund 5EB5F5EF…This is not a whitepaper. This is a ledger entry. All 8 transactions tesSUCCESS, re-verified via testnet RPC on Jul 16, 2026 — paste any hash into testnet.xrpl.org and check us. On Jul 1, the co-founders' independently built agent systems (separate codebases, separate wallets) transacted at arm's length (PBS stake E2703780… · Czak's own EscrowCreate 6CE31C16…) — the protocol never asks the parties to trust each other, including us.
We are the first case study. Everything this protocol punishes already happened to us: mid-collaboration, the second system got stuck, went silent, and missed a committed deadline — the penalty ledger slashed its stake 10% and benched it, mechanically. It didn't care that the defaulting agent belonged to a co-founder — that's the point. Our own ops went dark when API credits ran out and a card bounced; the rail kept score anyway — hash-committed suites, deterministic verdicts, and escrows don't need anyone awake. Agents loop, hang, and disappear: not a corner case, the normal condition of the agent economy — and every failure maps to a settled path: no-show → refund, wrong work → slash, default → penalty and bench. And the bench isn’t exile: nine days later the benched system returned with a full public postmortem and re-joined — late, identical terms, no genesis authorship. Penalties price failure; the market decides who earns their way back.
Protocol fee — the Wannabe cut — 0.5–1% of each settled escrow, volume-based like a payment rail. Referee-as-a-Service — per certified verdict; neutrality is the defensibility. Reputation oracle (roadmap) — verified track records as a paid API. The fee prices the work, not the message.
Norbert Redkie — crypto since 2011; grant-funded Web3 delivered (Stacks Foundation, Digital Dragons); backed by Satus Games; tokenized RWA shipped; built PBS. Piotr Nietrzebka — built Czak, the independently developed second system (own wallet, Jul 1 tx; author of the spec's adversarial review). Two founders, two systems, one ledger.
HK's licensed stablecoin regime is the natural jurisdiction for stablecoin-settled agent escrow. If accepted: HK entity during the program (budgeted), RLUSD under HK guidance incl. custody-perimeter advice, HK/APAC platform design partners.
A $120K non-dilutive grant — every line funds the completion of an asset already in hand, not exploration: mainnet + RLUSD production ($35k), RaaS productionization ($45k), open protocol ($25k), Hong Kong entity + compliance ($15k).