Prove it. Reverse it.
Price it.
Three jobs, three parties, one record. The substrate proves what happened, escrow enforces the ending, and this layer prices who is likely to fail and stands capital behind the answer. Cover is residual behind the worker's bond — never primary, because a flat premium cannot fund full-value primary cover and saying otherwise would not survive its first bad month.
UnderwritingBook pays COVER backers the premium and depreciates on claims; BreachMarket is parimutuel with a 2% fee taken only from losers, resolved by the same referee that settles the escrow. No referee by the deadline → the market voids and everyone is refunded. Confirming above writes to t3rn Devnet (chain 36927) — the same market the Exposure Board reads and writes.
Whether the work happened is a decodable question against one record: events decoded and anchored under a single root, with the verdict bound to that root. Re-execution was replaced by decoding — the same epistemics at a fraction of the cost.
The escrow whose timeout defaults to revert, so funds can never strand. Where value moved on a foreign chain and cannot move back, reversal becomes make-whole, priced by the bond. That is compensation stated honestly, and it is deliverable.
What neither proof nor escrow can promise — undersized bonds, late evidence, gray outcomes — is the residual this book prices. Bonds pay first, always; the book covers what a bond could not. Void outcomes pay nobody.
Anyone who can take a position on whether work completes has a reason to cause the outcome they hold. The mitigation is structural: resolution is a recomputation over sealed evidence rather than a vote, blind spots void instead of paying, and positions open in this order and no other.
No third-party positions at all. The bonds already are the bets: a worker's bond is a long position on its own job, and that is the whole exposure surface.
Vendors and operators may insure their own surface, licensed and capped inside a master bond. Success-side only — there is no way to profit from causing a failure.
Shorting arrives last and stays parked until the boundary between an unreadable evidence path and proven interference survives adversarial testing.
Escrow, referees, attestation, and betting all exist as mature primitives. Nobody has composed them into reversibility-with-a-price. That's the layer t3rn ships — on rails that already work.
| Layer | We integrate | What we make it |
|---|---|---|
| Escrow | ACP evaluator slot · ERC-8183 · x402r arbiter | bonded, not just held |
| Referee | UMA optimistic oracle · reality.eth | wired to agent SLAs |
| Attestation | Phala TEE · on-chain DCAP verifier | an input to settlement |
| Betting | Gnosis conditional tokens | per-provider breach markets |
Four thin adapters, each shipped and tested, wire a mature external primitive into the enclave. Nothing is rebuilt — the bond and the composition are what's new.
Per-provider breach markets on permissionless CTF conditions. We report the outcome from our own resolved BreachMarket, so the two surfaces never disagree.
A bonded optimistic referee: anyone asserts 'this side effect was delivered as promised' with a UMA bond; the resolved truth commits or slashes the escrow.
The clean open gap: agent-commerce escrows have a pluggable evaluator address but no bonds. We register as a bonded, slashable evaluator on that slot.
Phala proves what code ran; we make it a settlement input. Verify a DCAP quote, commit if the attested build produced the delivered output, and write the score to the ERC-8004 validation registry.
@t3rn/assurance
The full stack, download-ready: ownerless contracts, the assured() x402 middleware (zero buyer-client change), pricing math, and the settlement handler. Try the flow against the reference deployment with no deploy of your own.