Experiment

2. Pre-Flight Checklist

Required before designing the experiment:

• Single variable — changing ONLY one thing? If user wants to test multiple things → discourage, explain why
• No conflict — no other running experiments that would interfere?
• Stable baseline — do we have 2-3 weeks of baseline data? No current trend?
• Safe — checked interactions with current medications and conditions
• Measurable — do we know WHAT to measure? Do we have the tools (wearable, scale, etc.)?
• Evidence exists — is there reason to believe this could work?

If something isn't met → tell the user and suggest how to fix it.

3. Research Intervention

If we don't have research on this intervention:

• Run /research skill (or at least perplexity_ask for quick check)
• Determine: mechanism, dosing, expected timeline, safety

If we already have research:

• Load relevant file from docs/research/

4. Define Metrics & Baseline

Primary metric — what exactly to influence:

• Must be measurable (number, not feeling)
• Must be available (wearable, lab test, subjective 1-10 scale)
• Must have baseline (at least 2 weeks of data)

Secondary metrics — what else to watch:

• Side effects
• Confounders

Design selection:

Phase durations:

5. Fill Template

Use experiments/TEMPLATE.md and fill in:

• Experiment ID: next number in sequence
• Hypothesis: If [intervention], then [outcome] because [mechanism]
• Protocol: exact doses, timing, duration
• Metrics: primary + secondary
• Phases: dates for ABAB
• Pre-flight checklist: checked
• Safety review: interactions, stop criteria

Save as: experiments/EXP-XXX-short-name.md

6. Post-Setup

After creating the experiment:

1. Watchlist — add to active experiments section
2. Domain — add to Experiments section in relevant domain
3. Stack — if supplement, add to stack.md (with note "EXP-XXX")
4. Remind user:
   • When to start baseline measurements
   • What to track daily
   • When to come back for evaluation

Common Mistakes (remind the user!)

1. Regression to the mean — don't start during an extreme (very bad/good week)
2. Placebo effect — use objective metrics, not just "I feel better"
3. Confirmation bias — don't peek at data during experiment, evaluate at the end
4. Multiple variables — ONE change! If you want to test 3 things, design 3 experiments

If user provided an argument

/experiment cold exposure → design cold exposure experiment /experiment magnesium → design magnesium experiment /experiment → ask what they want to test