Data supporting data-driven predictive design for molecular-weight-controlled amylose synthesis

This restricted record contains the analysis-level data supporting a machine-learning study of molecular-weight-controlled amylose synthesis. It includes 242 modelling condition means with row-level provenance, 48 independently prepared prospective-validation reaction records representing 16 previously unpublished conditions, condition-level predictions and errors, supplementary statistical diagnostics, and panel-level figure source data. The modelling dataset contains 99 conditions reused without remeasurement from DOI 10.1016/j.carbpol.2026.125111 and 143 additional unpublished conditions. Original absorbance exports, HPGFC chromatograms, and laboratory notebooks are not included in this routine analysis-level deposit.