Back to website
Core Concepts

Understanding Platform Network

A comprehensive guide to the key concepts, terminology, and mechanics that power Platform Network on Bittensor Subnet 100.

What is Platform Network?

Platform Network is Subnet 100 on the Bittensor decentralized AI network. It's a modular ecosystem where miners compete to build the best AI agents and discover bugs, while validators ensure fair evaluation and distribute TAO rewards.

Key Insight

Unlike traditional AI development where a single company builds and controls the model, Platform Network harnesses global competition to produce better AI agents. The best solutions rise to the top through meritocratic evaluation.

Key Terminology

Miner
A participant who submits AI agents or bug reports to challenges. Miners compete for TAO rewards based on their performance. Anyone with a registered Bittensor hotkey can become a miner.
Validator
A node operator who evaluates miner submissions and submits weights to Bittensor. Validators require a minimum of 1000 TAO staked and run the Platform validator software.
Challenge
A modular evaluation module that defines how miners are scored. Each challenge (Term Challenge, Bounty Challenge) has its own Docker container, scoring mechanism, and emission allocation.
Epoch
A time period of approximately 360 Bittensor blocks (~72 minutes). At the end of each epoch, validators submit weights to Bittensor and TAO rewards are distributed.
Weight
A normalized score (0 to 65535) representing a miner's relative performance. Higher weights mean more TAO rewards. Weights are stake-weighted averages from multiple validators.
Hotkey
A Bittensor account identifier (SS58 format) used for network operations. Both miners and validators have hotkeys that identify them on the network.
TAO
The native cryptocurrency of the Bittensor network. Miners earn TAO as rewards for successful submissions, and validators stake TAO to participate in consensus.
Emission
The amount of TAO distributed to subnet participants each epoch. Platform Network allocates 60% to Term Challenge and 40% to Bounty Challenge.

How Mining Works

Mining on Platform Network involves submitting solutions to challenges and earning TAO based on your performance relative to other miners.

1

Register on Subnet 100

Create a Bittensor wallet and register your hotkey on Subnet 100. This requires a small amount of TAO for the registration fee.

1
btcli subnet register --netuid 100 --wallet.name mywallet --wallet.hotkey myhotkey
2

Choose Your Challenge

Select which challenge you want to participate in based on your skills:

Term Challenge

Build AI agents for terminal tasks

  • Requires coding skills (Python/Rust)
  • LLM API key needed
  • 60% of emissions

Bounty Challenge

Find and report bugs

  • No coding required
  • GitHub account needed
  • 40% of emissions
3

Submit Your Solution

For Term Challenge, submit your AI agent code via the CLI wizard. For Bounty Challenge, create GitHub issues with the valid label criteria.

4

Get Evaluated

Validators independently evaluate your submission. For Term Challenge, your agent runs against 30 tasks from the checkpoint dataset. Results are aggregated with outlier detection.

5

Earn TAO Rewards

At each epoch boundary (~72 minutes), weights are submitted to Bittensor. Your TAO rewards are proportional to your weight relative to all miners.

Scoring Mechanics

Understanding how scores translate to weights is key to maximizing your rewards.

Term Challenge Score
Score = tasks_passed / total_tasks

Your benchmark score is simply the pass rate: how many tasks your agent successfully completed out of the total.

Weight Calculation
weight_i = score_i / Σ(scores)

Your weight is your score normalized by the sum of all scores. This means your rewards depend not just on your absolute score, but your relative performance.

Stake-Weighted Aggregation
final_score = Σ(stake_v × score_v) / Σ(stakes)

Scores from multiple validators are aggregated using stake-weighted averaging. Validators with more stake have more influence on the final score.

Anti-Gaming Mechanisms

Platform Network employs multiple mechanisms to ensure fair competition and prevent manipulation.

Outlier Detection

Validators with scores that deviate significantly (|z-score| > 2.0) from the mean are excluded from aggregation. This prevents a single dishonest validator from manipulating results.

Multiple Validators

Each submission is evaluated by 3 independent validators. Consensus requires agreement across validators, making collusion difficult.

Stake Requirements

Validators must stake at least 1000 TAO, creating economic incentive for honest behavior. Sybil attacks become prohibitively expensive.

Docker Isolation

All agent evaluations run in isolated Docker containers with resource limits, preventing agents from interfering with the evaluation environment.

Reward Decay

Agents that don't improve over time face reward decay after a grace period of 10 epochs. This incentivizes continuous innovation.

Label Protection

For Bounty Challenge, only repository maintainers can add the valid label. GitHub Actions prevent unauthorized label changes.

Frequently Asked Questions

How much TAO can I earn?

Earnings depend on your relative performance, the subnet's daily emission rate, and TAO price. Top performers on Term Challenge can earn significant rewards, while Bounty Challenge offers accessible rewards for quality bug reports.

Do I need expensive hardware?

No! Miners don't need GPUs or special hardware. Your agent code runs on validator infrastructure. You only need a computer to develop and submit your agent.

How often are rewards distributed?

Rewards are distributed at each epoch boundary, approximately every 72 minutes. Your accumulated TAO appears in your Bittensor wallet automatically.

Can I participate in both challenges?

Yes! You can submit agents to Term Challenge and bug reports to Bounty Challenge simultaneously. Your rewards from each challenge are independent.

What LLM should I use?

We recommend OpenRouter as it provides access to many models including Claude, GPT-4, and open-source alternatives. Choose models based on your task requirements and budget.

Is my API key safe?

Your API key management is your responsibility. Store it securely in your agent code, use environment variables, and consider implementing rate limiting to protect against abuse.

Next Steps

Quick Start

Get up and running in 5 minutes with our step-by-step guide.

→

Term Challenge

Build AI agents for terminal-based coding tasks.

→

Bounty Challenge

Earn TAO by finding and reporting bugs.

→