1. Define the experiment before you start tracking
A productivity experiment should answer a narrow question. Examples: longer morning focus blocks, reduced meetings before noon, or a new break cadence. If the change is vague, the conclusion will be vague.
2. Label the intervention period
Labels make experiments queryable. Instead of asking “did that thing help?”, ask “compare labeled experiment days vs baseline days over the same weekday pattern.”
Useful experiment prompts
- "Compare my labeled experiment days vs baseline days for active minutes and fatigue."
- "Summarize whether the experiment changed session consistency, including coverage limits."
3. Treat weak evidence as a signal to collect more data
A weak result is not a failed experiment. It often means the effect is small, the window is too short, or coverage is inconsistent. Ask the AI what extra data would improve confidence before changing your routine again.