The consensus jury

Merit's premium high-assurance tier. Each answer is broken into atomic claims; only borderline claims escalate to a panel of models from different families — so their errors are uncorrelated, not five copies of one model agreeing on the same hallucination. A reputation-weighted supermajority decides each claim, and settlement is graded per claim (pay what passes, never all-or-nothing). Every ballot is TEE-attested and committed under a Merkle root.

Run a live panel with your API key →
Runs the real diverse-model panel on the 0G judge (forcePanel, $0 — no settlement). Your key is used only for this request and never stored.