At the very least there is a substantial burden of proof not just to show that each adjustment made could somehow be justified, but that no adjustments that would have confounded the point being demonstrated were omitted (and that’s a tough one to show, with UHI sitting there all naked and unaccounted for or unbelievably accounted for, and likely other adjustments to consider as well), and how it just so happens that when one adds up all of the corrections, they are in a perfect linear relationship with that primary variable. Correlation may not be causality, but with a correlation this good, you’d better be prepared to prove that it isn’t even inadvertently causal, because the point you want/need to make depends as much or more on the adjustment as it does anything in the unadjusted data!
That’s the killer, right? If half the TCS estimate goes away with unadjusted data, I’d argue that the lower bound of TCS is even smaller than the unadjusted estimate. You just don’t get to adjust and increase precision at the same time, at least not according to information theory or empirical practice. And “scientists” make this mistake all the time, which is why we are just now learning that eating fat and dietary cholesterol doesn’t increase your blood cholesterol, so bacon and ice cream and eggs — in moderation, as there is still a connection to total calories and obesity — is no longer a sin. That’s what happens to an entire branch of science, for as long as decades, once somebody sets out to prove a point and gets to select the data to use to do so.