This is the version of the explanation I wish somebody had given me, written down before I forget what confused me. It is about lean-mass preservation, and it is deliberately narrow — everything I am not confident about is marked as such.
What is actually established
There is also a quality change that the scan cannot see. Weight loss shifts fiber-type distribution toward Type I and improves mitochondrial density and insulin-stimulated glucose uptake per fiber — so the muscle you keep is metabolically better even though there is less of it. That does not make the loss irrelevant, but it explains why function often holds up better than the number suggests.
The condition it depends on
The condition that a single DEXA is not a measurement, it is one draw from a distribution. Hydration status moves the lean compartment by a kilogram or more between scans, so any comparison of scans done at different times of day or different hydration states is noise.
The practical version
If you are tracking this properly: same scanner, same time of day, same hydration, at least six months apart, and read grip strength or a lift number alongside it. Function is the outcome; the scan is the proxy.
What I am not sure about
The bit I cannot resolve on my own is what fraction of loss being lean mass is actually normal, because the numbers quoted here range from 10% to 40% and cannot all be right. Happy to be told the question itself is wrong.
— steph_laguna · corrections welcome and will be edited into this post with credit