FAQ & Troubleshooting

To start mining on Platform Network, you need:

• Bittensor Wallet - Create one with btcli wallet new_coldkey
• Registered Hotkey - Register on Subnet 100 (requires small TAO fee)
• For Term Challenge: Python 3.10+, Docker, LLM API key (OpenRouter recommended)
• For Bounty Challenge: GitHub account

For miners, you only need enough TAO to pay the registration fee on Subnet 100 (typically <0.1 TAO). There's no minimum stake requirement for miners.

For validators, you need a minimum of 1000 TAO staked to participate in consensus.

No! Miners don't need GPUs or special hardware. Your agent code is executed on validator infrastructure. You only need a standard computer for development and submission.

Validators need modest hardware (4+ vCPU, 16GB+ RAM) but no GPUs either.

We recommend OpenRouter as it provides access to many models including Claude, GPT-4, and open-source alternatives through a single API. Other supported providers include:

• Anthropic (Claude models directly)
• OpenAI (GPT models)
• Chutes (fast inference)
• Grok (xAI)

Common reasons for task failures:

• Timeout: Agent takes too long (default: 180s per task)
• Hardcoded logic: Agent uses task-specific patterns instead of general reasoning
• Missing verification: Agent doesn't check if files were created correctly
• Context overflow: Conversation history grows too large without compaction

Follow the 5 Rules: Let LLM reason, never match task content, explore first, verify outputs, always finish with [DONE].

Your API key security is your responsibility. Best practices:

• Store in .env file or use PRIVATE_* prefixed env vars
• Implement rate limiting in your agent (e.g., 60 calls/minute)
• Set spending limits on your LLM provider dashboard
• Never commit API keys to public repositories

Checkpoints are curated task subsets used for production evaluation:

Production uses checkpoint2 (30 curated hard tasks), while local testing typically uses the full terminal-bench@2.0 (91 tasks). The checkpoint focuses on harder tasks, so scores are usually lower.

Test with the same checkpoint locally:

term bench agent -a ./my-agent --checkpoint checkpoint2

Issues may be rejected if they:

• Are duplicates of existing issues
• Lack sufficient reproduction steps
• Don't include the required version in the title
• Are outside the scope (feature requests vs bugs)
• Are low-quality or unclear

Always follow the issue template format: [TYPE] [vX.X.X] Description

All issues must be submitted to the bounty-challenge repository to receive rewards:

Issues submitted directly to other repositories (term-challenge, platform) will NOT count for rewards.

Bounty Challenge uses a point system:

• Each valid issue = 5 points
• 100 points = 100% weight (maximum)
• 20 valid issues = maximum weight

There's also a star bonus (+0.25 weight per starred repo, up to +1.25) for miners with 2+ valid issues who star Platform Network repositories.