design-product-experiment

Installation
SKILL.md

Design Product Experiment

Contract

Design backward from the decision. Test the riskiest claim with the cheapest method that can answer it. Do not use an A/B test when observation, interview, prototype, smoke test, or technical probe is more appropriate.

Workflow

  1. State the decision and falsifiable hypothesis: population, change, expected behavior, mechanism, and time window.
  2. Identify the assumption type: value, demand, comprehension, usability, feasibility, viability, or causal impact.
  3. Choose the method and comparison. Define assignment unit, eligibility, control, exposure, contamination risks, and novelty effects.
  4. Select one primary decision metric with exact calculation and window. Add diagnostic and guardrail metrics.
  5. Estimate sample and duration from baseline, minimum meaningful effect, variance, power, seasonality, and traffic; use ranges when inputs are unknown.
  6. Predefine exclusions, stopping rules, data-quality checks, segment analysis, and interpretation for positive, negative, null, or inconclusive results.
  7. Define the operational plan, owner, ethics or privacy checks, and the decision each result triggers.

BuildBetter Acceleration

If BuildBetter MCP is available, search organization skills first. Use signals, survey responses, people or account context, documents, and synthetic persona tests to refine hypotheses. Synthetic results can reveal questions or usability risks but cannot establish real-world causal impact.

Installs
1
GitHub Stars
4
First Seen
Oct 27, 2026