model-evaluation

This plugin provides comprehensive model evaluation beyond accuracy, including sklearn classification metrics with bootstrap confidence intervals, probability calibration analysis, sliced evaluation by subgroup, fairness metrics (demographic parity, equalized odds), and LLM output quality assessment using BERTScore and RAGAS. It includes an agent expert in evaluation patterns, reusable code skills for calibration and statistical testing, and a command-line workflow for generating full evaluation reports, fairness analysis, and model comparisons.