Bittensor Brain Subnet

A Bittensor subnet focused on validating prediction market results. This subnet leverages open-source verification methods to determine the truth value of statements with confidence intervals.

Overview

The Brain subnet is designed to create a decentralized prediction market verification system where:

• Miners verify statements by determining if they are true or false with confidence intervals
• Validators evaluate miners' verification results for accuracy and proper methodology

This creates an incentive mechanism where the most accurate and reliable verification miners receive a greater share of TAO emissions.

Architecture

The subnet consists of the following components:

Core Components

• Protocol: Defines the communication protocol between miners and validators
• Miners: Verify statements using open-source methods and provide results
• Validators: Evaluate miner results and distribute rewards based on accuracy

Key Features

• Statement verification with confidence intervals
• Evidence-based verification methodology
• Cross-validation to prevent cheating
• Reward mechanism based on accuracy and evidence quality

Installation

Prerequisites

• Python 3.8+
• Bittensor

Install from Source

# Clone the repository
git clone https://github.com/yourusername/bittensor-brain-subnet.git
cd bittensor-brain-subnet

# Install the package
pip install -e .

Usage

Running a Miner

To run a miner on the Brain subnet:

# Run a miner with default settings
brain-miner --netuid 1 --wallet.name miner --wallet.hotkey default

# Run with custom settings
brain-miner --netuid 1 --wallet.name miner --wallet.hotkey default --miner.verify_timeout 120

Running a Validator

To run a validator on the Brain subnet:

# Run a validator with default settings
brain-validator --netuid 1 --wallet.name validator --wallet.hotkey default

# Run with custom settings
brain-validator --netuid 1 --wallet.name validator --wallet.hotkey default --validator.run_validation

How It Works

Verification Process

1. Validators select statements from a database to be verified
2. Miners receive statements and perform verification using:
   • Web searches
   • Database lookups
   • Logical reasoning
   • Evidence collection
3. Miners return results with:
   • True/False determination
   • Confidence score (0.0-1.0)
   • Supporting evidence
   • Explanation of reasoning
   • Methodology description
4. Validators evaluate results based on:
   • Accuracy (compared to ground truth when available)
   • Evidence quality
   • Explanation quality
   • Methodology soundness
5. Validators distribute rewards to miners based on performance

Validation Process

To prevent cheating, validators may also request miners to validate other miners' results:

1. Validators send a miner's verification result to other miners
2. These miners evaluate if the verification was performed correctly
3. If issues are detected, they provide an alternative result
4. This cross-validation helps ensure integrity in the network

Development

Project Structure

bittensor-brain-subnet/
├── brain/                  # Core subnet functionality
│   ├── __init__.py         # Package initialization
│   ├── protocol.py         # Communication protocol definitions
│   ├── forward.py          # Forward pass implementation
│   └── reward.py           # Reward mechanism implementation
├── neurons/                # Neuron implementations
│   ├── miner.py            # Miner implementation
│   └── validator.py        # Validator implementation
├── docs/                   # Documentation
├── setup.py                # Package setup
└── README.md               # Project documentation

Customizing the Miner

To create a custom miner with specialized verification methods:

1. Inherit from the base BrainMiner class
2. Override the perform_verification method with your custom logic
3. Implement specialized search and analysis methods

Example:

from neurons.miner import BrainMiner

class CustomBrainMiner(BrainMiner):
    def perform_verification(self, statement):
        # Your custom verification logic here
        # ...
        return is_true, confidence, evidence, explanation

License

This project is licensed under the MIT License - see the LICENSE file for details.