A problem with a post-hoc adjustment imo is that it makes the connection between the professor/material and the grade received more confusing. Students typically want to know: what does the professor expect, and what do I have to do to get an A (or a B)? Halfway through the semester, a struggling student might want to know, am I on track to fail (or get a D perhaps), and is it plausible for me to bring it up or should I drop the course? The professor might have specific suggestions for improvement, and give rough feedback on what the student needs to do for the rest of the course to get at least a C (or maybe even a B).

But if the answer is that the professor themselves isn't quite sure, because it'll go through some statistical adjustment process after the course ends, that's a bit less helpful.

I'm admittedly thinking more about smaller courses and especially project- or seminar-based ones. Large lecture courses with homework+tests that are just numerically graded and averaged could more plausibly be done as he proposes.