Information-Guided Selectivity in Viral Host-Cell Targeting: A Quantitative, Falsifiable Account of Apparent Viral Purpose

Viral host-cell targeting is routinely described in intentional language: viruses "seek," "choose," or "target" particular cells. We ask what empirical content, if any, licenses this language and answer without appeal to consciousness, literal purpose, or self-replication as a criterion of life. The defensible core is information-guided selectivity: productive viral entry is statistically distinguishable from passive, availability-weighted encounter, and this departure is measurable in bits using standard information theory and binding thermodynamics. We formalize selectivity as a Kullback-Leibler divergence between realized entry and receptor availability, derive it from a two-state mass-action binding scheme, and establish its basic mathematical properties (non-negativity, a sharp zero under a non-selective null, and consistency of a plug-in estimator with an explicit finite-sample power analysis for a proposed single-cell assay). Parameterizing this measure from published affinity data for two of the best-characterized viral targeting systems, namely SARS-CoV-2 spike binding across ACE2 host orthologs, and influenza hemagglutinin binding α2,6 versus α2,3 sialosides, and propagating the reported affinity ranges by Monte Carlo (5×10^4 draws, fixed seed), we obtain robustly positive selectivities of 1.42 bits (95% CI [1.06, 1.95]) and 0.31 bits (95% CI [0.19, 0.45]) respectively, with realized engagement concentrated overwhelmingly on the preferred targets. The measure is identically zero under the non-selective null by construction, giving a sharp, falsifiable operationalization of "non-random targeting." We further compute the integrated information of a deliberately minimal capsid-receptor state-transition motif using the whole-minus-sum effective-information measure, obtaining a modest 0.75 bits; this exceeds every member of a matched random-network null ensemble (n=3000; null mean 0.32±0.06 bits, maximum 0.573 bits), yet behaves non-monotonically under lesion (rising to 0.91 bits, a known general property of whole-minus-sum measures rather than an anomaly of this motif), and we report both facts honestly rather than treating Φ as a fitness proxy. Embedding a bounded selectivity term in a standard Susceptible-Infected-Recovered (SIR) model raises the simulated epidemic peak by 40% under an illustrative coupling; a reproducible multi-method global sensitivity analysis (Latin-hypercube/PRCC and variance-based Sobol indices, with an explicit interaction decomposition) quantifies each parameter's contribution and shows selectivity to be a dependable but secondary driver relative to the classical epidemiological rate constants. Throughout, "agency" and "intentionality" are used strictly as functional shorthand, with their metaphysical commitments explicitly bounded and contrasted against the ongoing falsifiability debate surrounding Integrated Information Theory. All numerical results, tables, and figures are produced by a single self-contained, seeded Python script reproduced in the appendix, so that every reported value in the main text is exactly reproducible from that script.