Guided Stochastic Assembly: A Scaffold-and-Fill Theory of Thought

Conventional accounts of neural coding treat trial-to-trial variability as noise to be averaged away, and treat the repeatable, phase-locked signal that survives averaging as the message. This article proposes the reverse division of labor. In Guided Stochastic Assembly (GSA), a thought is neither the clean signal nor the noise. It is the pattern that emerges when low-dimensional, temporally precise guiding dynamics — oscillatory phase, traveling waves, top-down prediction, and neuromodulatory gain — constrain high-dimensional, stochastic population activity. The guide supplies timing, routing and a scaffold of expectations. The stochastic activity supplies candidate content: a rapidly sampled set of possible assemblies. Synaptic weights, learned context and current goals then select which candidate stabilizes, and stabilization — not the guide, and not the raw variability — is the thought. The article first reviews the established foundations the theory builds on: predictive coding, communication-through-coherence, theta–gamma phase-amplitude coupling, neural sampling, stochastic resonance, criticality, and ephaptic coupling. It then states GSA as four distinct functional roles — guide, fill, selection, stabilization — and formalizes them as coupled dynamics from which quantitative predictions follow. Six hypotheses are advanced, each with a pre-stated falsification condition. Chief among them: a decoder that reads guiding phase and stochastic population activity jointly will decode an imagined action more accurately than either alone, and the advantage will track trial-wise phase–amplitude coupling. A complete proposed study is specified to test the theory: imagined and performed hand actions, including the multi-step action open the can, recorded with high-density EEG and simultaneous muscle and eye controls; five decoder families compared head-to-head; and artifact controls — phantom-head recordings, sham phase, and label shuffling — designed to defeat the theory if its apparent fill signal is in fact artifact. An interpretation matrix commits in advance to what every outcome pattern, including a full null, would mean. The article closes with the theory's limits, the ethics and privacy obligations attaching to any technology that can decode intended action, and a sober assessment of BCI and AI collectors: devices can stimulate at brain-rhythm frequencies and implanted systems can decode intended speech and movement clinically, but a passive external collector of thought content is not supported by present evidence — and GSA, if true, makes decoding harder and more personal, not easier and more universal.