Proof-of-Quality
PoQ is an open protocol for verifiable, consensus-based quality signals in AI workflows. Add a verifiable quality layer to existing AI workflows without replacing tools or processes.
Human-Powered AI Verification
Don’t trust, verify. PoQ adds a verifiable quality signal where subjective judgment matters.
Why PoQ
AI systems fail not because models are weak, but because subjective quality is poorly enforced. Without a quality signal, you get drift, errors and confident responses that are not reliable.
The onchain world offers two rules that help in this environment: don't trust, verify, and make sure incentives push people toward honest behaviour. PoQ brings those ideas into AI with staking, local consensus, onchain attestations and built-in data sovereignty.
PoQ produces a trusted signal. Your system decides what to do with it.
Core Concepts
Oracles
Connection between PoQ and external systems used by validators/contributors. They handle interaction with tools, agents, and workflows.
Validation
Independent judgments are collected and weighted by stake to reach local consensus.
Attestation
A verifiable onchain record that downstream systems can audit and trust. Immutable provenance for training and RL loops.
Protocol Roles
Originator creates tasks. Contributor does the work. Validator reviews submissions. Adapters connect PoQ to external tools.
Data Sovereignty
Your data stays in your bucket. PoQ only verifies quality without requiring you to hand over control.
Local Consensus
Validator committees reach agreement on quality. Aligned validators earn rewards, others are slashed.
Who PoQ is For
PoQ is designed for builders who need enforced quality, not just output evaluation.
Agent builders (LangChain, LangSmith, Onchain Agents, custom stacks)
Teams building autonomous or semi-autonomous agents where actions have real cost.
- •Need a confidence signal before executing high-stakes actions
- •Want quality enforcement without hard-coding rules
- •Running agents in production where failures are expensive or irreversible
RLHF & HITL Pipeline Operators
Teams managing reinforcement learning or human-in-the-loop systems at scale.
- •Collect subjective judgments that must be reliable, not noisy
- •Require consensus instead of single-reviewer approval
- •Want auditability and provenance for training and evaluation loops
Tool Builders & Platform Integrators
Teams building tools, frameworks, or platforms used by AI developers.
- •Want to embed quality as a first-class primitive in their product
- •Can extend PoQ through adapters instead of bespoke integrations
- •Benefit from shared infrastructure rather than rebuilding QC systems
ML & Deep Learning Model Builders
Training models that require high-quality labeled datasets with provable annotation quality.
- •No cryptographic proof of annotation correctness
- •Manual spot-checking doesn't scale across datasets
- •Need verifiable quality guarantees for model training
- •CVAT lacks economic incentives for quality
System Architecture
Application Layer
Create or validate tasks with a simple user interface.
Oracles
Adapters act as oracles between PoQ and external systems. They handle interaction with tools, agents, and workflows, while PoQ focuses on producing a verifiable quality signal.
Contributor Oracle
The Contributor Oracle interfaces with the system where work is performed. It:
- • Connects to external tools (annotation tools, agent frameworks, pipelines)
- • Collects task outputs or actions for evaluation
- • Submits work into PoQ for quality assessment
This oracle ensures PoQ can evaluate work without controlling how or where it is produced.
Validator Oracle
The Validator Oracle interfaces with reviewers and validators. It:
- • Presents work for independent evaluation
- • Collects quality judgments
- • Submits evaluations to PoQ for consensus and scoring
Validator oracles do not modify the original tool or workflow. They report judgments only.
Protocol Core
Task Management
Task Registry • Funding • Rewards
Economic Layer
Staking • Slashing • Identity
Validator Committees
• Stake-weighting • Aligned validators earn rewards, others get slashed
Storage
S3 • Google Cloud Storage • IPFS
Your data stays with you
Blockchain
L2 Base • Ethereum Attestation Service
Onchain provenance & trust
What You Get
Ready to Build with PoQ?
Whether you're building adapters, integrating PoQ into your AI pipeline, or exploring new use cases - let's discuss how we can help.
Book a Call