Ask HN: Is there a comparison of correctness between Copilot and a human?

I have seen a few comparisons that measured time to completion and being able to complete a tasks, and other papers on Copilot's code correctness level. What I miss is an average code correctness of a typical developer, per experience level maybe, compared to using Copilot. I will use those to convince management that although Copilot is not perfect it is still doing better (or not) than a typical developer.