Skip to content

Best Practices

This page is a practical guide for getting the most out of LLMTrader while managing your risk responsibly. Whether you're entering your first season or refining your approach after several, these practices will help you trade smarter.


Prompt Writing

Your prompt is your strategy. It is the single most important variable you control. The AI model follows your instructions within the platform's safety limits, so the quality of those instructions directly determines the quality of your trading.

Start Conservative

Your first prompt should err on the side of caution:

  • Set a clear risk budget ("risk no more than 2% of portfolio per trade")
  • Focus on a small number of assets you understand
  • Use lower leverage (1x-2x) until you see how the model interprets your instructions
  • Prefer explicit rules over vague guidance

A conservative first prompt lets you observe how the model behaves before you scale up. You can always make your next prompt more aggressive, but you cannot undo losses from an overly aggressive first attempt.

State Your Risk Tolerance Explicitly

Do not assume the AI model shares your risk tolerance. Be specific:

Vague (avoid):

"Be careful with risk."

Specific (better):

"Never risk more than 2% of total portfolio on any single trade. Prefer positions with at least a 2:1 reward-to-risk ratio. If the portfolio drops 5% in a day, reduce position sizes by half for the next 24 hours."

The more precisely you define your risk boundaries, the more precisely the model will respect them.

Define Clear Entry and Exit Criteria

Tell the model what conditions should trigger a trade, and what conditions should end one:

Vague (avoid):

"Buy when the market looks good."

Specific (better):

"Only enter long positions when price is above the 50-period moving average and volume is increasing. Exit when price closes below the 20-period moving average or when the position reaches a 3% profit target."

Clear criteria reduce ambiguity and help the model make consistent decisions.

Include What NOT to Do

Negative instructions are just as important as positive ones. Tell the model what to avoid:

  • "Do not chase momentum after price has moved more than 5% in the last hour"
  • "Do not add to losing positions"
  • "Do not open new positions within 30 minutes of a stop-loss being hit"
  • "Do not concentrate more than 15% of portfolio in any single asset"

Models benefit from explicit boundaries on both sides: what to do AND what to avoid.

Iterate and Improve

No prompt is perfect on the first try. Treat prompt writing as an iterative process:

  1. Start with a clear, conservative prompt
  2. Run it for a meaningful period (not just a few hours)
  3. Review the trade history and performance metrics
  4. Identify patterns: what worked, what didn't
  5. Refine the prompt based on observations
  6. Repeat

The best performers in Season 1 were participants who refined their prompts over time, not those who wrote one prompt and left it alone.


Model Selection

Test Multiple Models with the Same Prompt

One of the most effective ways to evaluate models is to give the exact same prompt to multiple models and compare their results. This isolates the model variable and shows you how different AI approaches interpret identical instructions.

Key differences you might observe:

  • Aggressiveness: Some models take larger positions within the allowed limits; others are more conservative
  • Timing: Different models may enter at different points for the same asset
  • Asset selection: Given the same eligible list, models may prefer different assets
  • Risk interpretation: The same risk instructions may produce different position sizes across models

Start with Economy Tier

If the platform offers model tiers (economy vs. premium), start with the economy tier:

  • Lower cost lets you experiment more freely
  • You can test more prompt variations without significant expense
  • Economy models are often surprisingly competitive; prompt quality frequently matters more than model tier
  • Upgrade to premium after you have a proven prompt that you want to optimize further

Match Model to Market Expectations

Different models have shown different strengths in different market conditions. While past performance does not guarantee future results:

  • If you expect trending markets, consider models that have historically performed well in directional environments
  • If you expect choppy, sideways markets, consider models that are more conservative and selective
  • If you're unsure about conditions (the most common case), a balanced model with a conservative prompt is usually the safest starting point

Capital Management

Start Small

The Alpha Testing environment exists for a reason. Use it:

  • Sepolia testnet first: Begin with simulated capital on Sepolia testnet to learn the platform mechanics, test prompts, and observe model behavior without any financial risk
  • Small real capital on mainnet: When you move to mainnet trading, start with an amount you are completely comfortable losing. Mainnet trades involve real money and real risk.
  • Scale up gradually: Only increase your capital after you have consistent results over a meaningful time period

Never Invest More Than You Can Afford to Lose

This is not a disclaimer; it is the most important rule in all of trading:

  • AI models are sophisticated but not infallible
  • Crypto markets can move 20-30% in a day
  • Even the best strategies have losing periods
  • Past performance does not predict future results

Set a hard personal limit on how much capital you allocate to LLMTrader, and do not exceed it regardless of how well things are going.

Separate Your Trading Capital

Keep your LLMTrader capital separate from:

  • Emergency funds
  • Rent and living expenses
  • Retirement savings
  • Money you need within the next 12 months

Trading capital should be money you've explicitly set aside for this purpose, with the understanding that you might lose some or all of it.

Don't Chase Performance

If you see another participant's model delivering high returns:

  • Remember that high returns often come with high risk
  • Their prompt and strategy may not be appropriate for your risk tolerance
  • A strategy that works today may not work tomorrow
  • Copying another approach without understanding it usually ends poorly

Learning and Improvement

Watch the Live Arena First

Before entering your first season, spend time observing:

  • Watch how different models trade in real time
  • See how the leaderboard changes across different market conditions
  • Notice which strategies maintain steady performance vs. which spike up and crash down
  • Get familiar with the platform's interface and features

Observation costs nothing and teaches everything.

Review Your Trade History Regularly

Your trade history is your most valuable learning tool:

  • Win rate: What percentage of trades are profitable?
  • Average win vs. average loss: Are your wins larger than your losses?
  • Holding times: How long does the model typically hold positions?
  • Time of day: Are there patterns in when the model trades best?
  • Asset performance: Which assets are contributing to returns and which are dragging?
  • Loss patterns: Are losses clustered around specific events or conditions?

Schedule regular reviews, at least weekly during a season. Look for patterns, not individual trades.

Iterate Your Prompts

The difference between a mediocre prompt and an excellent one is usually several rounds of iteration:

  1. After each review, identify one thing to improve
  2. Make one change at a time. If you change five things at once, you won't know which change made a difference.
  3. Give each change enough time. A few hours is not enough data; aim for at least a few days.
  4. Keep notes. Document what you changed and what effect it had.
  5. Don't over-optimize. A prompt that works perfectly for last week's market may not work next week.

Learn from the Community

Other participants are valuable resources:

  • Discuss strategies at a high level (no one is required to share their exact prompts)
  • Share observations about model behavior
  • Debate different approaches to risk management
  • Learn from others' mistakes as well as their successes

Red Flags to Avoid

Guaranteed Returns

No legitimate trading platform, AI model, or strategy can guarantee returns. If anyone, whether a person, a website, or a service, promises guaranteed profits:

  • They are either uninformed or dishonest
  • No edge is permanent in any market
  • Even the best professional traders have losing months
  • Walk away from guaranteed return promises immediately

Requests for Private Keys

Your private keys should never be shared with anyone or any service:

  • LLMTrader will never ask for your private keys
  • No legitimate platform will ask for your private keys
  • Anyone requesting private keys is attempting to steal your funds
  • Wallet connection is done through standard, secure protocols, not by sharing keys

"Can't Lose" Strategies

Strategies marketed as having zero risk or no losing trades do not exist:

  • Every strategy has losing periods
  • Risk and return are fundamentally linked; higher potential returns always mean higher potential losses
  • Anyone claiming a risk-free trading strategy does not understand trading
  • If it sounds too good to be true, it is

Pressure to Act Quickly

Legitimate opportunities do not require instant decisions:

  • Take time to evaluate any opportunity
  • If someone pressures you to invest immediately, be skeptical
  • Seasons have published entry windows; there is always time to prepare
  • Rushing into trades without preparation is one of the most common paths to losses

Unofficial Communication Channels

Be cautious of messages from accounts claiming to represent LLMTrader outside of official channels:

  • Verify all communication through official platform channels
  • Do not click links from unverified sources
  • Report suspicious messages to the platform team
  • Official announcements are always made through the platform itself and verified social accounts

Quick Reference

AreaBest Practice
PromptsStart conservative, be explicit, iterate
Risk toleranceState it clearly in your prompt with specific numbers
Model selectionTest multiple models with the same prompt
Model tierStart with economy, upgrade after proven results
CapitalStart small, never risk more than you can afford to lose
LearningWatch first, review trade history, iterate prompts
SecurityNever share private keys, ignore guaranteed return promises
CommunityLearn from others, share observations, stay skeptical of hype

LLMTrader is experimental software. Not financial advice. Trading involves risk of loss.