The Limits of Falsifiability: Dimensionality, Measurement Thresholds, and the Sub-Landauer Domain in Biological Systems

This paper examines when a biological experiment leaves rival explanations observationally indistinguishable. The public interpretation is protocol-relative: underdetermination under one measurement map is a reason to design a new intervention, not to declare biology outside falsification.

The paper identifies three levels of limitation: (1) physical detection limits set by noise, sampling, backaction, and the detector; (2) observation-map loss, where an assay may discard the distinction on which rival models differ; and (3) framework dependence through auxiliary assumptions. Landauer's bound is not a universal signal-detection threshold, and a one-bit decision need not reconstruct an organism's full state to be sufficient for a targeted hypothesis.

Controlled physical systems often permit isolation, symmetry, and repeatability that biology does not. Both fields also contain difficult inverse and identifiability problems. The difference is methodological and task-specific, not a boundary where mathematics stops working.

The practical response is to state the measurement map, identify observational equivalence classes, and seek interventions on which rival mechanisms predict different outcomes.