ÍRIS Evidence Base: Evaluation Strategies for LLM Outputs

This dataset contains the evidence base used in the development of the ÍRIS taxonomy, a multidimensional taxonomy and decision-support approach for the selection of evaluation strategies for LLM outputs. The dataset comprises evidence extracted from 47 primary studies published between 2023 and April 2026. Two complementary files are provided. The first contains the initial article-level extraction, with one record per study. The second contains the standardized evidence base, in which the extracted information was decomposed into individual evidence records representing specific relationships between evaluation scenarios and strategies. The data include information about application domain, task, quality aspect, evaluation objective, automation level, risk, evaluation methods and metrics, and standardized evaluation families. The standardized dataset comprises 179 evidence records and was used to support the construction and empirical analysis of the ÍRIS taxonomy.