Model A scores two points above B. What does that mean? The items, prompts, grading rules and statistical methods can all change what that score means.
I organized these questions into a 20-chapter model evaluation tutorial, starting with grading one item and moving through data and protocols, graders, model judges, paired comparisons and recomputable reports.
It includes five interactive labs and Python examples, with complete English and Chinese versions. All example data is synthetic teaching material, not measured model results.
Read the complete model evaluation tutorial →
You can also download the complete offline package for both languages, code and teaching data.