Show Training Data Was Checked
Prove that domain experts reviewed your training datasets against explicit criteria instead of blindly trusting a vendor’s word.
Verifiable human judgment for high-stakes data. Independent expert panels score labeling, AI agent outputs, and compliance audits against custom rubrics—sealed as auditable onchain records.
Any time one person’s opinion isn’t enough, Proof of Quality coordinates an independent panel of experts and turns their consensus into a provable score.
Prove that domain experts reviewed your training datasets against explicit criteria instead of blindly trusting a vendor’s word.
Benchmark model responses and multi-step AI agent trajectories against strict behavioral rules using a consensus group rather than a single grader.
Convert security reviews, vulnerability flags, and compliance checks into permanent, cryptographically signed records customers can verify.
Swap inconsistent, one-person sampling for structured panel agreement on every critical item across production workflows.
Switch between core task tracks to inspect real evaluation workflows, rubric checks, and estimated payouts.
Define what good looks like, let an independent expert panel score every item, and receive a signed report anyone can verify on Base.
AI teams bring data and define the exact scoring rubric. Contributors start at Trainee or higher tiers and stake SAPIEN collateral where required.
A panel of independent reviewers scores each submission against the rubric. Consensus builds contributor reputation from Trainee to Contributor, Expert, and Master.
Agreed results are sealed as a permanent, auditable record on Base Mainnet—unlocking instant rewards for accurate contributors and verifiable provenance for teams.
Every completed evaluation answers who reviewed an item and whether the review can be trusted—backed by smart contracts on Base Mainnet.
// Sapien PoQ Attestation Record (Base Mainnet)
{
"network": "base-mainnet",
"rubricId": "poq-agent-safety-v2",
"panelQuorum": "5/5 Independent Experts",
"consensusScore": 0.98,
"status": "SEALED_ONCHAIN"
}
How contributors describe task pacing, rubric consistency, and peer-validated payouts across the network.
“Pairwise response and agent step checks are straightforward once you read the rubric carefully. Writing a concise, specific reason keeps my consensus score high.”
“Occlusion and boundary boxes on street and 3D scenes reward patience over rushing. Moving up reputation tiers opened cleaner batches and better reward rates.”
“Checking claims against cited sources catches subtle hallucinations that look convincing at first glance. Staking and panel consensus keep standards honest.”
Compatible with major Web3 wallets on Base & EVM