Growth Experimentation
Boundary
Own the learning loop, not external platform configuration. Do not call a winner from inadequate evidence or test deceptive claims, hidden material terms, false urgency, forced continuity, exploitative targeting, discriminatory treatment, or unnecessary personal-data collection.
Required Inputs
- Decision and business outcome
- Audience, channel, control, and candidate variable
- Baseline, sample availability, cycle length, and instrumentation
- Primary metric, guardrails, cost, decision stakes, privacy limits, and protected audience constraints
Workflow
- Write: if X changes for Y, Z should change because R.
- Choose one primary variable and document unavoidable differences.
- Define control, variants, assignment, measurement, sample, duration, and contamination risks.
- Set a minimum meaningful effect and decision rules appropriate to volume and stakes.
- Run without opportunistic peeking or changing the outcome definition.
- Classify as keep, discard, or inconclusive and explain uncertainty.
- Promote durable findings into a playbook with scope, caveats, and a retest trigger.
Output
- Experiment specification
- Instrumentation and risk checklist
- Decision log
- Playbook update
- Ranked next experiments by learning value
Quality Gate
- The metric connects to business value.
- Guardrails protect economics, user choice, privacy, accessibility, and safety.
- Variants disclose material terms and never depend on deception, involuntary enrollment, or exploitative segmentation.
- Causal language matches the design.
- Low-volume findings remain provisional.
1---2name: growth-experimentation3description: Run marketing experiments with explicit hypotheses, isolated variables, decision rules, guardrails, and a living evidence playbook. Use when campaigns or content variants need a keep, discard, or inconclusive decision instead of vanity reporting.4---56# Growth Experimentation78## Boundary910Own the learning loop, not external platform configuration. Do not call a winner from inadequate evidence or test deceptive claims, hidden material terms, false urgency, forced continuity, exploitative targeting, discriminatory treatment, or unnecessary personal-data collection.1112## Required Inputs1314- Decision and business outcome15- Audience, channel, control, and candidate variable16- Baseline, sample availability, cycle length, and instrumentation17- Primary metric, guardrails, cost, decision stakes, privacy limits, and protected audience constraints1819## Workflow20211. Write: if X changes for Y, Z should change because R.222. Choose one primary variable and document unavoidable differences.233. Define control, variants, assignment, measurement, sample, duration, and contamination risks.244. Set a minimum meaningful effect and decision rules appropriate to volume and stakes.255. Run without opportunistic peeking or changing the outcome definition.266. Classify as keep, discard, or inconclusive and explain uncertainty.277. Promote durable findings into a playbook with scope, caveats, and a retest trigger.2829## Output30311. **Experiment specification**322. **Instrumentation and risk checklist**333. **Decision log**344. **Playbook update**355. **Ranked next experiments by learning value**3637## Quality Gate3839- The metric connects to business value.40- Guardrails protect economics, user choice, privacy, accessibility, and safety.41- Variants disclose material terms and never depend on deception, involuntary enrollment, or exploitative segmentation.42- Causal language matches the design.43- Low-volume findings remain provisional.
majesticlabs-dev