What problem does it solve? Marketers running paid-ad experiments often lack a rigorous way to plan sample sizes, isolate variables, and judge whether a result is statistically and practically significant before acting on it. This Skill produces a falsifiable hypothesis, variant matrix, power plan, and a documented effect/uncertainty read-out from your own exported test data. ## Core Features & Use Cases - Experiment Design: Builds a hypothesis, one-variable-per-variant matrix, primary/secondary/guardrail metrics, and a sample-size, duration, and power plan for creative, landing-page, and incrementality tests. - Statistical Read-out: Applies two-proportion z-tests, Mann-Whitney U, or bootstrap confidence intervals to your exported results CSV, reporting statistical and practical-effect flags separately. - Decision Governance: Applies only a precommitted, owner-approved action rule; otherwise returns decision UNDECIDED so a p-value never silently becomes a business action. - Use Case: You have a finished test results CSV with per-variant sessions and conversions and need to know whether the winner is significant and whether to promote or kill it. ## Quick Start Design an A/B test for two landing-page hero variants with a 3% baseline CVR where I want to detect a 15% lift.
