Measuring what Matters: Construct Validity in Large Language Model Benchmarks

Measuring what Matters is a systematic review of construct validity practices in benchmarks for LLMs. This repository contains the code and data from the review.