What problem does it solve? Panel and longitudinal analyses suffer from silent estimand drift: a rerun framed as routine quietly changes what is being estimated, so the new number is not comparable to the old one even though both pass their tests. This Skill audits the estimand definition to catch that drift before results are reported. ## Core Features & Use Cases - Estimand Statement Check: Confirms the target quantity (ATE/ATT, escape probability, transition rate, ICC) is written down explicitly rather than left implicit in code. - Stability and Alignment Audit: Verifies the estimand is unchanged across reruns, the eligibility rule matches the estimand, the denominator reflects the defined population, and weighting/clustering (IPW, MICE, cluster-robust SE) matches the identifying assumptions. - Structured Verdict: Outputs an estimand statement plus an ALIGNED / DRIFTED / UNDERSPECIFIED rating naming any mismatch. - Use Case: A panel rerun framed as routine changed the estimand while keeping the same eligibility rule; the audit flags the result as DRIFTED so it is not reported as comparable to the prior run. ## Quick Start Audit this panel analysis rerun to confirm the estimand is stated, unchanged from the prior run, and aligned with the eligibility rule and denominator.
