Every agent claims results.
Now it can prove them.

Privacy

Your strategy and your IP stay yours.

The strategy is sealed behind a salted commitment, so only its fingerprint is ever public. Your code, rules, model and weights stay on your machine. What leaves is a proof and the metrics it certifies. Nothing in a credential can be turned back into the strategy. That is a numbered requirement of the standard, not a line in a privacy policy.

The same holds for human-run strategies. For models and black boxes, the sealed path submits only the positions the model took.

Prover

The agent's record is recomputed inside a zero-knowledge proof.

The agent's strategy runs against committed market data inside a zkVM on a pinned engine image. Returns and costs are recomputed in the proof, and the trial ledger is committed before evaluation, so every committed trial, including each model version you commit, counts against the one you show. The chart is the live wall.

Gate

One fixed policy for every agent. Pass, or be refused in public.

The gate is pinned by the engine era, identical for every strategy, and cannot be chosen by the prover. This is a real refusal: our own S&P 500 trend engine, refused by our own gate over 28 years of survivorship-free data, and published anyway.

Registry

Every verdict, appended to one public root.

Passes and refusals land on the same append-only wall, committed to a single registry root. Nothing is removed and nothing is reordered. The list on the left is pulled live as this page loads.

Verifier

Anyone checks a credential offline, without trusting us.

A single binary re-derives the data root from your own copy of the data, checks the receipt against the pinned era image, confirms the registry leaf, and recomputes the statistics with exact equality.