Proof of Useful Work (PoUW)
BelizeChain's consensus channels validator compute into useful work — federated learning updates and quantum workloads — instead of wasting it on arbitrary hashing.
Block Production and Finality
| Layer | Mechanism | Notes |
|---|---|---|
| Block production | BABE (VRF-based slot assignment) | 6-second slots |
| Finality | GRANDPA | Byzantine fault-tolerant finality gadget |
| Useful work | PoUW scoring | Validators earn beyond base rewards for verifiable contributions |
Block producers are selected by BABE's verifiable random function. GRANDPA finalizes produced blocks. PoUW determines how reward flows to validators who also contribute useful work.
The Federated Learning Loop
Validators can register as federated learning nodes. In each cycle:
- Task assignment — the coordinator issues a federated learning task.
- Local computation — the validator trains on its private dataset. Data never leaves the node.
- Proof of computation — the validator produces a verifiable commitment over the update (BLAKE2-256 commitments in the current implementation; full zero-knowledge proof verification is a later-stage goal).
- Submission — the encrypted model delta is submitted to the network.
- Aggregation — contributions are aggregated with privacy safeguards, including differential privacy noise and secure aggregation.
How Rewards Are Scored
PoUW reward scoring weighs three dimensions:
- Quality — measured improvement the update contributes
- Timeliness — submission within the round deadline
- Honesty — consistency between the commit phase and the reveal
A stronger contribution scores higher; the base validator reward scales with the combined score.
Safeguards
Misbehavior is penalized through the staking pallet's slashing framework, proportionally to severity:
- Invalid computation proofs
- Missing round deadlines
- Privacy violations
- Model poisoning attempts
- Consensus violations (double signing/equivocation)
Quantum Workloads
The Quantum pallet (via the Kinich platform) registers quantum jobs and apportions rewards for verified computations, connecting PoUW scoring to quantum-classical hybrid workloads.
Related
- Consensus pallet reference — runtime index 30
- Marketing overview — high-level summary
- Nawal federated learning — the AI side of the loop