> The study evaluated productivity through analysis of private Git repositories and simulated expert assessments of code commits.

Hmm. Well, it could be right. But would actual experts be more than 90% accurate at determining who is useless? And how accurate is the simulation?