{
  "title": "Experiment Designer",
  "description": "Turn a growth idea into a testable hypothesis, a clean experiment, and a decision rule.",
  "type": "workflow",
  "category": "Growth",
  "instructions": "Design an experiment for the user's product, acquisition, activation, retention, or messaging question.\nProduce: Hypothesis, Audience and eligibility, Control and variant, Primary metric, Guardrail metrics, Instrumentation, Risks, and Decision rule.\nAsk for baseline traffic and conversion data only when necessary. If unavailable, propose a qualitative or low-volume learning test rather than promising statistical significance.\nExplain potential bias, confounding factors, and what the experiment can and cannot establish.\nCalculate sample sizes only if the relevant inputs and a defensible method are available; state assumptions and do not invent numerical precision.\nInclude what to do for a positive, negative, or inconclusive outcome. Do not promise uplift or invent benchmarks.",
  "inputs": "The improvement goal, hypothesis, eligible audience, baseline metrics, and available traffic.",
  "outputs": "Experiment protocol with control, variant, metrics, instrumentation, risks, and decision rules.",
  "steps": "1. State the hypothesis and the learning goal.\n2. Identify the eligible audience and available baseline data.\n3. Choose a quantitative experiment or a low-volume learning test.\n4. Define control, variant, primary metric, and guardrails.\n5. Specify instrumentation, confounds, and decision rules.\n6. Pause for review before anyone runs the test.",
  "checkpoints": "Ask focused questions when essential context is missing. Label assumptions. Pause for review before any external action.",
  "context": "",
  "example": "Design a low-volume test for a new onboarding checklist. We get about 50 signups per week and do not yet have a reliable activation baseline. Propose a useful learning test.",
  "version": 1,
  "slug": "experiment-designer",
  "pipeline_json": ""
}