I did read the thread. I'm objecting to the idea that stacking lossless corrections onto lossy compression and then measuring the total is a good way to measure what we want to measure here, wrt human knowledge. It may be the best we have, but it's not good.
Why should we care what you think though? Im not being nasty, but unless you have a reputation in that field you have to give some cogent argument or its just some random possibly-uninformed opinion.
If your goal is compressing human knowledge, then you do want to avoid wasting bits on the details of wording that were random chance.
The problem is inability to objectively judge such a compression, not the mere fact that it won't be bit-perfect.
It is not "not even wrong".