Three minds. One verified answer.

Every answer is produced by three independent AI systems from different providers — arranged to disagree, and forbidden to be polite.

Drafts the answer as numbered, checkable claims — every claim tagged with its source.

Independently attacks every claim and must name its own evidence. Agreement without a source is rejected as invalid.

A third system, blind to both identities, rules on citations alone — and may rule a claim unresolved rather than force a winner.

The final answer is assembled from accepted claims only.

Built to resist confident nonsense

Blind stages — no system knows which model produced the text it is judging.
Evidence-gated — agreement counts only with an independently named source.
Honest uncertainty — unverified claims are dropped or flagged for you, never smoothed over.
Continuously audited — the pipeline is regularly tested against planted falsehoods and its catch rate is monitored.

Physical actions are never authorized by consensus alone — a human confirms every step.

Claim → audit → verdict

Invitation-only access. Tell us about your process and hazards, and we'll respond directly.