What problem does it solve? Confirmatory analyses lose credibility when specifications are chosen after seeing results. This Skill prevents specification searching and p-hacking by locking the hypotheses, estimand, primary specification, sample rules, robustness suite, and decision rule before any outcome data is touched. ## Core Features & Use Cases - Pre-committed analysis lock: Fixes six elements — directional hypotheses with multiple-comparisons correction, the exact estimand, one primary specification, sample and exclusion rules, a small targeted robustness suite, and a decision rule with power/MDE commitments. - Blinding gate with sign-off: Persists the plan to a file (e.g., docs/pre-analysis-plan.md) and requires explicit user sign-off before outcome data is loaded, plotted, or inspected. - Confirmatory vs. exploratory separation: Keeps a bright line between pre-registered claims and exploratory findings, routing deviations through a checkpoint rather than silent judgment calls. - Use Case: Before reading out an A/B test or policy evaluation, use this Skill to write down the primary metric, model, sample rules, and what would count as a null result — then get sign-off before unblinding the outcomes. ## Quick Start Before we look at any results, help me write a pre-analysis plan that locks the hypothesis, primary specification, sample rules, and decision rule for this experiment.
