Serial decoding systems converge to a throughput band of 3–6 bits per event across genetics,cognition, language, and AI. We derive a discrimination cost function W(M, ε) with super-linearalphabet scaling and exponential fidelity scaling, and show the basin is geometrically inevitablefor any system with super-linear discrimination cost (α > 1). We identify two physically distinctcost regimes. Regime A (Alphabet-Bound): biological systems using pairwise molecularrecognition, where cost scales with the number of discriminated symbols (αalphabet ≈ 2, fromM(M−1)/2 combinatorics), producing a thermodynamically constrained basin. Regime B(Capacity-Bound): silicon AI using independent learned parameters, where cost scalessub-linearly with model capacity (αcapacity = 0.937, measured across 27 models), andvocabulary is a cheap linear projection—producing no cost-driven basin. The biologicaltemperature prediction (thermophiles show reduced amino acid entropy) receives preliminarysupport from Kazusa-verified data (partial r = −0.451, p = 0.014, n = 29, pending phylogeneticcontrols). The framework explains why the biological basin exists, and why silicon escapes itthrough architectural decoupling of vocabulary from discrimination cost.
Grant Lavell Whitmer III (Fri,) studied this question.