Toward Trustworthy Measurement: A Scoping Review of Evaluation Methods for Large Language Models

A PRISMA-ScR scoping review mapping evaluation methodology for large language models. The review synthesizes three literatures that have largely developed in parallel: benchmark construction, meta-evaluation of automated judges, and measurement validity. It charts the evaluation method families in use, the scoring and validity-control techniques applied, and the challenges that limit the trustworthiness of reported model comparisons. Search window 1 January 2018 to 31 March 2025 across eight information sources. This project holds the registered protocol, the search query log, and the PRISMA-ScR reporting checklist.