Tests should include pulling indexed work from OpenAlex and matching the metadata with the pipeline's predictions.