[ppeetteerr]

They are addressed but the core of the thesis is still wrong:

> This is the core problem: our entire evaluation infrastructure is structurally reactive. We measure the system after it has changed. We never predict the change.

That's kind of the point of evals.