Working sheet / 01

Experiment release check

Use this before sharing a model result. Checked items do not guarantee validity; retain the underlying evidence.

Tick each item after you have retained the supporting notes in your own records. This page stores nothing; refreshing clears the ticks. Print a copy if you need to keep it.

• The prediction time and target are written down.
• Repeated people, devices or accounts cannot leak across the split.
• Preprocessing is fitted using training data only.
• We compared with a simple baseline on the same split.
• The final test was kept separate from tuning.
• We retained the data version, seed, settings and failure examples.

0 of 6 questions reviewed

After the review

Record the unanswered questions, name an owner for each and set a review date. Do not treat an unchecked item as a failure or a checked item as independent assurance. The notes behind the decision matter.