What problem does it solve? Experimental results in ML/AI research are often summarized with cherry-picked best scores, missing significance tests, or decorative figures. This Skill enforces an evidence-first workflow that validates experiment artifacts, runs proper descriptive and inferential statistics, and produces real publication-quality figures with explicit interpretation. ## Core Features & Use Cases - Strict statistical analysis: Computes mean ± std, 95% confidence intervals, significance tests with assumption checks, effect sizes, and multiple-comparison corrections, with non-parametric fallbacks when assumptions fail. - Real scientific figures: Generates actual comparison, ablation, and training-dynamics figures from readable data, each with purpose, caption requirements, and interpretation checklists. - Structured analysis bundle: Produces analysis-report.md, stats-appendix.md, figure-catalog.md, and a figures/ directory, explicitly separating evidence from manuscript prose. - Use Case: You have seed-level CSV results comparing a full fine-tuned model, an adapter variant, and a frozen-encoder baseline. This Skill validates the comparison units, runs paired tests with Holm correction, and outputs a complete analysis bundle ready to hand off to a results-report writer. ## Quick Start Ask the assistant to run a strict statistical analysis of the experiment results in your results directory and generate the full analysis bundle with figures.
