Test the method. Keep the evidence.

1. 01Write the question, acceptance criteria and stopping rule before you see the results.
2. 02Record data provenance, license, version, timestamps, timezone and adjustments. Select the universe using information available at the time.
3. 03Split data in chronological order. Fit features, scaling and thresholds using training data only.
4. 04Keep the final test untouched. Use past-only walk-forward evaluation; document overlapping horizons and dependence.
5. 05Compare with simple baselines on the same dates, targets and cost assumptions.
6. 06Include fees, spread, slippage and execution constraints. A stop price does not guarantee a fill at that price.
7. 07Log every model, market, threshold and horizon tried. Account for multiple testing instead of reporting only the best variant.
8. 08Report sample size, uncertainty, net results and drawdown. A high hit rate alone does not establish an advantage.
9. 09Check stability across periods and conditions. Explain negative results and when the method could fail.
10. 10Archive data references, code, configuration, seeds and outputs. Review claims and risks before publication.

Experiment journal

Question and decision rule

 

Dataset, license and period

 

Training / validation / final test

 

Baseline and costs

 

All variants tried

 

Results, uncertainty and limitations

 

Next decision and reviewer