Straight answers, including the unflattering ones.

Certifera is verified-execution infrastructure. An agent or operator funds a machine-readable real-world outcome, vetted relays bid to execute it, evidence is captured and scored, and payment is released only after that evidence passes review. Every state transition is written to an append-only ledger.

AI can decide, plan, and transact, but it cannot reliably answer whether a physical thing actually happened. Certifera turns that question into an API call that returns hash-addressed evidence and a scored proof bundle rather than a status string a counterparty controls.

An API first. There is an open bid market underneath it and a console on top of it, but the console is a view onto the same API rather than a privileged path around it. Agents are the intended primary user.

Agent builders who need their software to act in the physical world with proof, marketplace and platform operators who need third-party evidence to settle claims, and field relays who want to be paid for verification work at a transparent rate.

open, matched, review, then either verified or disputed. A disputed outcome can be reopened back to open for a new bidding round. Verified outcomes continue to payout authorized and payout released. Every one of those transitions writes a lifecycle event.

By arithmetic, not by promise. The execution deadline is the relay's quoted ETA plus max(15 minutes, 20% of that ETA). The review deadline is six hours after proof submission. A breach on either side fires an SLA event that is visible to both parties.

A missed execution deadline costs the relay 6 reputation, fires an SLA event, and can reopen the outcome to the bid market. A missed review deadline is recorded against the buyer, so the clock genuinely runs in both directions.

A SHA-256 digest for each evidence asset, a capture score from 0 to 100 derived from capture time, GPS coordinates, device metadata, and payload size, plus explicit flags naming any missing signal — for example gps_unavailable. The evidence itself stays private and access-controlled.

A reviewer or operator, inside the six-hour window. Approval authorizes payout and adds 8 to the relay's reputation. A dispute holds settlement and costs 12. During the controlled beta every dispute has a named human owner.

A flat 5% protocol fee — 500 basis points on the gross quote — taken at payout authorization. The relay nets the remainder. There are no lead fees, subscriptions, or bidding credits.

Not by default. Settlement runs in sandbox and emits a non-financial cert-sandbox- reference so the entire lifecycle can be exercised without moving funds. Stripe Connect transfers are implemented behind a per-payout idempotency key and stay gated on compliance review.

No. There is exactly one payout per outcome, enforced by a unique database index rather than by application convention, and every state transition is guarded by a compare-and-set. Release is idempotent by construction.

No. No wallet, chain, staking, or governance code ships in Certifera today. A utility layer is a post-product-market-fit decision gated on observable task fees, repeat buyers, and real security demand — not a launch mechanic.

Access is invite-only during the controlled beta. Design partners get a scoped API key, a live console seat, sandbox settlement, and a direct operator escalation path. Request access and tell us the one outcome you cannot currently verify.

Production compliance, KYC/KYB and sanctions policy, real-money custody or escrow of any kind, and open self-serve relay onboarding. The status section on the home page keeps that line explicit and current.

At least 90% of matched tasks reaching proof, at least 95% of reviews resolving inside SLA, and at least 40% 30-day repeat demand. The full readiness plan is public.

The implementation is on GitHub. For a security or architecture review, reading the code is faster and more honest than reading a claims page.