experiment-readout

Write experiment readouts for product, growth, pricing, UX, and operational tests. Use when Codex is asked to summarize an A/B test, pilot, experiment result, metric movement, guardrails, decision, or

SKILL.md

GitHub

Experiment Readout

Core Workflow

  1. Identify hypothesis, variants, audience, dates, primary metric, guardrails, sample, and decision rule.
  2. Summarize results against the pre-stated decision rule.
  3. Separate observed movement, statistical/practical significance, data quality, and interpretation.
  4. Check guardrails and unintended effects before recommending rollout.
  5. State decision: ship, iterate, stop, extend, or rerun.
  6. Capture learning, follow-up questions, and next experiment.

Safety Rules

  • Do not invent sample sizes, statistical significance, confidence intervals, or metric values.
  • Do not recommend rollout when guardrails failed or data quality is poor.
  • Escalate pricing, medical, legal, employment, financial, or high-risk customer-impacting experiments.

Deliverable Shape

For experiment readouts, provide:

  • Hypothesis and setup
  • Metrics and decision rule
  • Results summary
  • Guardrail review
  • Data quality caveats
  • Decision
  • Learning
  • Follow-up actions

References

  • Read references/experiment-readout-checklist.md when writing experiment or pilot readouts.