I have read his spreadsheets, so here's my critique. (And I will say that, throwing all the spreadsheets out the window, I still agree broadly with many points to some extent.)

The input numbers (in meta.xlsx) come from imperfect samples of the real world with non-zero error -- often quite a bit higher than 0 -- which is compounded pretty far in some cases. I don't think any person actually can actually evaluate these studies, which is a major problem with this entire idea and is a fundamental problem with any overly-quantitative solutions to problems.