Data sets the ceiling before the model does
Many AI trading products show models, equity curves, and polished interfaces, but do not explain where their data comes from. That is a problem. A trading model cannot be more reliable than the data it consumes. Is price delayed? Is news duplicated? Is volume abnormal? Does the backtest leak future information? These questions matter more than the model name.
Projects such as OpenBB and FinGPT are useful references because they separate financial data, text, models, and research workflows instead of only claiming that AI is smart.
Minimum data pipeline
Data source
→ cleaning
→ feature building
→ model / rule
→ risk gate
→ order decision
→ execution log
→ post-trade review
Every layer should be traceable. Timestamps, data source, model input, model output, risk rejection reason, order result, slippage, and final P&L should be logged. Without logs, there is no review. Without review, AI is just generating stories.
Data checklist
| Question | Why it matters |
|---|---|
| Is the data source reliable? | Bad prices create bad orders |
| Is there latency? | Event and fast trading are sensitive |
| Are duplicate news items cleaned? | Repetition can amplify false signals |
| Is there look-ahead leakage? | Backtests become inflated |
| Are failed orders recorded? | Successful-only logs beautify results |
| Can logs be exported? | Auditing and review require evidence |
Check Yourself
Why should you ask about data sources before trusting an AI trading product’s P&L screenshot?
Suggested answer: Screenshots can be selected or polished. Data sources, cleaning, latency, and logs determine whether the strategy is verifiable.
Further reading: OpenBB GitHub · FinGPT GitHub · Machine Learning for Trading · NIST AI RMF
Get the pre-trade checklist.
We are turning these guides into a searchable checklist for checking terms, rules and risk before you trade.