Before the run

  • Write one question and one decision you will make from the result.
  • Save the baseline, candidate configuration and repository commit.
  • Freeze input files; record a SHA-256 checksum and retrieval time.
  • Record when each observation occurred and when it became available to the decision.
  • Define missing data, stale data, fees and retry handling in advance.
  • Separate exploratory examples from a held-out evaluation.

During the run

  • Record the exact command, dependency versions and random seed where relevant.
  • Keep excluded observations and reasons alongside admitted observations.
  • Compare both rules on the same frozen evidence and cost assumptions.
  • Keep failures and uncertain outcomes visible. A timeout is not evidence that an external action failed.

Before sharing

  • Re-run from a clean environment.
  • Ask another person to follow the command without undocumented steps.
  • Mark synthetic inputs and worked examples clearly.
  • Report sample size, exclusions, costs and limitations alongside the result.
  • Keep secrets, account identifiers and private customer data out of the package.

Fill in your experiment

Question: ________

Commit and configuration: ________

Input file and checksum: ________

Decision-time availability rule: ________

Baseline and candidate: ________

Held-out window: ________

Run command: ________

Observed result and limitations: ________

This checklist supports engineering reproducibility. It does not establish trading profitability or predict an investment outcome.