What problem does it solve? Analysis results produced by code or research workflows can contain hallucinated numbers, fabricated citations, weak statistical methodology, or unverifiable claims. This Skill audits those results before they are trusted downstream, deciding whether to accept them or send them back for revision. ## Core Features & Use Cases - Source Reliability Hard Gate: Detects hallucinated statistics, fake references, and code-data misalignment, triggering immediate revision when fabrication is found. - Multi-Dimensional Scoring: Scores Accuracy, Completeness, Robustness, Relevance, Methodology, and Critical Reflection on a 0-10 scale with PASS/FAIL thresholds per dimension. - Statistical Methodology Audit: Checks multiple-testing correction, model assumptions, confounder control, sample size, batch effects, missing data, and reproducibility, with domain-specific priorities for biology, chemistry, materials, and finance. - Use Case: After a code-engineer agent produces a correlation analysis with r and p-values, run this Skill to verify the numbers are reproducible from the supplied code and data, audit the statistical assumptions, and receive an ACCEPT_AND_PROCEED or REVISE_AND_RETRY verdict with prioritized revision guidance. ## Quick Start Evaluate the analysis results just produced by code-engineer and give me an ACCEPT_AND_PROCEED or REVISE_AND_RETRY decision with dimension scores and revision guidance.
