A big 30-year Danish study that seems to be one of the main ones referenced for this purpose sure appears to be doing exactly what I worried it was: measuring the predictive power of BMI for mortality, not the healthiest weight for someone to maintain, which are different things. I see nothing in the abstract & methods about correcting for sick people tending to lose weight, and the healthiest BMI creeping up over time as mean population BMI increased, plus the U-shaped mortality curve, really do look like what you’d expect to see if they’re not accounting for that.