The paper with the non-algorithmic screening used synthetic resumes. I rather suspect that they didn't generate a realistic distribution of qualifications levels.