Governance Overview
NeuroShard uses a decentralized governance system to manage protocol upgrades. Any changes to the LLM architecture, training algorithms, or economics must go through a formal proposal and voting process.
Why Governance Matters
In a decentralized AI network, the model and economics are tightly coupled:
| Component | Economic Impact |
|---|---|
| Training algorithm | Determines reward efficiency |
| Model architecture | Affects hardware requirements |
| Inference speed | Impacts market pricing |
| Layer distribution | Influences node earnings |
Changing one component without adjusting others can:
- Inflate or deflate NEURO earnings unfairly
- Exclude nodes that don't meet new requirements
- Break verification mechanisms
Governance ensures all stakeholders have a voice in these decisions.
Core Principles
1. Transparency
All proposed changes are public. Anyone can review the technical specification and economic impact before voting.
2. Economic Parity
Every proposal must include an Economic Impact Analysis that quantifies how earnings change.
3. Stake-Weighted Voting
Voting power is proportional to staked NEURO. Those with skin in the game make decisions.
4. Grace Periods
Approved changes include upgrade windows so nodes have time to adapt.
The NEP Process
NEP = NeuroShard Enhancement Proposal
┌──────────────────────────────────────────────────────────────────┐
│ NEP LIFECYCLE │
├──────────────────────────────────────────────────────────────────┤
│ │
│ ┌─────┐ ┌────────┐ ┌────────┐ ┌──────────┐ ┌──────┐│
│ │DRAFT│───►│ REVIEW │───►│ VOTING │───►│SCHEDULED │───►│ACTIVE││
│ └─────┘ └────────┘ └────────┘ └──────────┘ └──────┘│
│ │ │ │ │ │ │
│ │ │ │ │ │ │
│ Author 7 days 7 days Activation Applied │
│ submits technical stake-weighted block to │
│ review vote set network │
│ │
└───────────────────────────────────────────────────────────────────┘NEP Types
| Type | Code | Description |
|---|---|---|
| Architecture | NEP-ARCH | Model changes (attention, layers, embeddings) |
| Economics | NEP-ECON | Reward rates, fees, staking parameters |
| Training | NEP-TRAIN | DiLoCo params, gradient handling, aggregation |
| Network | NEP-NET | P2P protocol, gossip, routing |
| Governance | NEP-GOV | Changes to governance itself |
| Emergency | NEP-EMERG | Critical security patches (fast-track) |
Quick Links
Governance at a Glance
┌─────────────────────────────────────────────────────────────┐
│ GOVERNANCE FLOW │
├─────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────┐ │
│ │ PROPOSER │ Stake: 100+ NEURO │
│ │ (Any Node) │ Fee: 10 NEURO (burned) │
│ └──────┬──────┘ │
│ │ │
│ ▼ │
│ ┌─────────────┐ │
│ │ NEP │ Title, Motivation, Specification │
│ │ PROPOSAL │ Parameter Changes, Economic Impact │
│ └──────┬──────┘ │
│ │ │
│ ▼ │
│ ┌─────────────┐ │
│ │ VOTING │ 1 NEURO staked = 1 vote │
│ │ (7 days) │ 66% approval, 20% quorum │
│ └──────┬──────┘ │
│ │ │
│ ┌────┴────┐ │
│ ▼ ▼ │
│ ┌───────┐ ┌────────┐ │
│ │APPROVE│ │ REJECT │ │
│ └───┬───┘ └────────┘ │
│ │ │
│ ▼ │
│ ┌───────────┐ │
│ │ SCHEDULED │ Grace period (7-30 days) │
│ │ │ Nodes upgrade │
│ └─────┬─────┘ │
│ │ │
│ ▼ │
│ ┌───────────┐ │
│ │ ACTIVE │ New parameters enforced │
│ │ │ Protocol version bumped │
│ └───────────┘ │
│ │
└─────────────────────────────────────────────────────────────┘Example: Adding Multi-Token Prediction
Here's how a major training change would be proposed:
nep = create_proposal(
title="Add Multi-Token Prediction Training",
nep_type=NEPType.TRAINING,
economic_impact=EconomicImpact(
training_efficiency_multiplier=2.0, # 2x faster training
training_reward_multiplier=1.0, # Same reward per batch
net_earnings_change_percent=0.0, # Neutral (quality gains)
),
upgrade_path=UpgradePath(
grace_period_days=14,
backward_compatible=True,
),
)The economic impact shows:
- Training becomes 2x more efficient
- Per-batch rewards stay the same
- Net effect: Model improves faster, making tokens more valuable
This ensures miners aren't suddenly earning half as much while the network benefits from faster progress.