The judgment
layer for AI output.
Vetted reviewers correct, fact-check, and sign off on what AI produces — before it reaches a customer, a patient, a court, or a reader. Open to companies and individuals alike.
Three steps, one accountable trail
Every task moves through the same sequence, so every output has a traceable, timestamped record of who checked it and what changed.
A company submits AI output
Model output, confidence score, and context are attached — a labeled dataset, a drafted decision, generated content, a flagged transaction.
A matched reviewer corrects it
Reviewers are matched by domain credential and task type, not just availability. They mark up, correct, or reject the output inline.
The correction is signed and logged
A timestamped, attributable sign-off is attached to the record — the artifact your compliance or ops team can point to later.
Open review tasks
Priced by domain and disagreement risk — not a flat per-item rate.
Real people, certified and rated
Every reviewer is a real, credentialed person — a lawyer, a pharmacist, a specialist. Stars come from the people whose work they've actually reviewed.
Priced by what the model got wrong before
Disagreement-weighted pricing
A flat per-task rate pays a careful reviewer and a careless one the same. Payout here scales with domain risk and the model's historical error rate on similar tasks — reviewers earn more for catching what the AI actually tends to miss.
- BASE Set by task complexity and domain
- RISK+ Higher payout where past disagreement rates run high
- STREAK Consistent accuracy raises a reviewer's standing rate over time