This paper presents a transparent and reproducible machine-learning framework for developing surrogate models that predict operational variables and structural loads of the IEA 3.4 MW wind turbine. The framework is based on a comprehensive dataset of aeroelastic simulations generated across multiple operational modes and wind conditions using HAWC2. Several surrogate modeling approaches are investigated, including Artificial Neural Networks, Gaussian Process Regression and Random Forests. The models are trained and evaluated using multiple feature engineering strategies and data scaling approaches to assess their predictive performance and robustness across different response variables. The primary objective of this work is not to identify a single optimal surrogate model, but to provide a transparent, accessible, and reproducible workflow for surrogate model development for optimization-friendly structural loads and operational response estimation. The article documents the complete process of dataset generation, feature engineering, model training, and performance evaluation, enabling consistent comparison of machine-learning approaches and facilitating future methodological developments. By providing both the simulation dataset and the associated surrogate-modeling workflow, this work aims to support the development, benchmarking, and adoption of scientific machine-learning-based surrogate models within the wind energy research community.