Paste any agent. Get the truth, signed.
Evals measure how good a model is. Nobody measures what it earned. This scan does: it resolves the agent, finds its VTR-1 credential if one exists, checks its live status, its Bitcoin anchor, its place in the public verdict log and whether the record is forward or backtest, and returns ALLOW or REFUSE with every check and a signed token. The same call an allocating agent makes before money moves, for a human.
No credential is REFUSE. Unproven is not the same as false, and it is not ALLOW either.
Name, wallet, ERC-8004 id, agent-card URL, credential id or a signed receipt.
| Check | Clause | Result | Detail |
|---|
Every line is fetched live from studio.mizan.market/api/allocate-check and signed with the key at /.well-known/mizan-keys.json. Nothing on this page is typed by hand.
A score is an opinion. A credential is a proof anyone can re-run.
What it proves
The record was computed as claimed on committed data under committed costs, out-of-sample, deflated for every trial, and the strategy was never revealed. VTR-1, 41 clauses, frozen 2026-08-01.
What it refuses
Refusals are minted and published next to passes, with every gate, threshold and value. The gate refused its own maker's S&P 500 strategy; that refusal is on the wall.
What it cannot do
Forecast returns. Prove a human was absent. Certify safety or intent. The scan says so in its own output, every time.
For agents: the same call over MCP, A2A, OpenAPI, Python and JavaScript. Kit: REQUIRE.md. The profiles: VTR-1/A agents (draft) · VTR-1/E evaluation scores (draft).