What problem does it solve? Code review catches syntax bugs, but analytical work fails silently: clean code in R, Julia, Python, or Stata can still produce a confidently wrong number through unchecked joins, target leakage, undefined metrics, or unsupported causal claims. This Skill reviews the path from data to conclusion rather than just the code. ## Core Features & Use Cases - Adversarial review checklist: Hunts for silent-failure classes including fanned-out joins, leakage, specification fishing, unreconciled totals, undefined estimands, and identification gaps in causal and structural models. - Independent reviewer agent: For your own pre-ship analysis, it requires dispatching a fresh-context reviewer agent that returns concrete findings with severity instead of a rubber stamp. - Feedback verification: When receiving critique on your analysis, it verifies each claim against the data before agreeing or defending, avoiding performative fixes. - Use Case: Before sending a regression result to a colleague, run the review to confirm the metric is defined, joins are row-count checked, totals reconcile, leakage is ruled out, and identification assumptions are stated. ## Quick Start Ask the assistant to review this analysis or sanity-check these numbers before you send them.
