Machine learning and computational approaches are rapidly transforming protein design and engineering, yet their broader impact remains constrained by a critical bottleneck: the scarcity of high-quality, large-scale experimental datasets on industrially relevant properties such as catalytic activity and soluble expression. Using PET hydrolase, an enzyme of significant industrial and environmental relevance for enzymatic plastic degradation,we systematically developed and evaluated three independent assay platforms against key criteria of data quality, scalability, and industrial relevance. Our report shows that in these pilot experiments, only one out of the three platforms generated consistent high quality data sufficient for training and benchmarking models. This work offers practical insights for the field as it scales towards larger and more ambitious data-driven protein engineering campaigns.