Choice Without a Central Chooser: A Falsifiable Multiscale Model of Neural Route Selection

Abstract

Neural tissue contains no single executive that inspects every available action, yet an organism can maintain alternatives, commit to one action, monitor its consequences, and alter later behavior. This paper proposes a multiscale model in which learned synaptic and dendritic states make cellular transitions criterion-sensitive; excitation, inhibition, and disinhibition change time-varying effective connectivity; distributed cortical and subcortical dynamics transform competing action-related states; and action-conditioned sensory feedback updates the criteria governing later transitions. Four biological scales are kept distinct: cellular transition, circuit route selection, population commitment, and organism action. The account credits prior work on recurrent path diversion, criterial causation, evidence accumulation, affordance competition, context-dependent dynamics, basal-ganglia and superior-colliculus selection, thalamic switching, cerebellar evidence processing, dendritic events, corollary discharge, and reafference. Its candidate contribution is the joined, testable ladder from learned criteria to trial-specific route, commitment, controlled action, consequence, and later updating. Draft 5 closes the primary-page locator for the paper's retained Tse atoms, states the influence, extension, and philosophical departure without absorbing criterial causation into SAN, and freezes an analysis plan for a prospective route screen in the International Brain Laboratory Brain Wide Map release. The screen is observational: it can reject or prioritize route candidates but cannot establish inhibitory causation, organism-level agency, or the complete multiscale ladder. The draft preserves the preparation-bounded subcortical comparisons, internal adversarial ledger, Buzsaki Figure 11.5 boundary, 2017 protocol-search genealogy, and deterministic synthetic fixture. Across 30 replicates per scenario, a prospective true-route variable improved held-out log loss by 0.0352 and passed the declared gate in 30/30 runs. Route-null and global-gain fixtures passed in 0/30. A post-outcome leakage variable produced an apparent gain of 0.5525 but was rejected in 30/30 because its provenance was invalid. An added timing feature was removed in 30/30 timing-null runs. These are software checks, not biological results. Receiver-relative phase, duration, waveform, magnitude, and spatial relations remain optional and must outperform anatomy, rate, power, movement, arousal, common input, and capacity-matched recurrent baselines. The proposal does not establish phenomenal consciousness, metaphysical freedom, moral responsibility, or one universal neural circuit.