Describe the observable rule
Specify the market, timeframe and exactly what is known when a decision is made. Use completed candles if the rule depends on their closing values. Avoid asking an AI to maximize a headline return without defining how the result should be evaluated.
Inspect a historical test first
Review the generated rule and actual historical trades. Record fees, slippage, fill timing and the tested date range. Check one entry and exit manually against the rule. A paper bot will not fix a misunderstanding embedded in the strategy.
Freeze an experiment version
Keep the code and settings that start the paper run. Define the observation horizon and operational stop conditions. A stop for stale data is different from a rule to exit a simulated position. Keep both explicit.
Review new information
At the review date, compare expected signals with recorded decisions. Inspect missed signals, restarts and periods without trades. If the strategy changes, preserve the old record and begin a new version. Do not claim that a short successful run establishes future returns.
Sources and next steps
- Premiss pricing and paper trading ↗Current free and paid allowances and the distinction from approved live access.
- Investor.gov — investment tools ↗Investor education and the importance of understanding investment risk.