Tomographic Coherence-Based Neural Rendering: A Falsifiable Neuroscience Protocol with Firewalled Bridges to Pattern, Action, Consciousness Science, and Physics

A brain preparing a grip does not receive one finished description of its own state. Different electrodes, cells, and moments expose partial measurements. The scientific problem is whether relations among those partial measurements contain a reproducible, task-relevant structure that a downstream system could use, beyond what is already available in voltage, power, firing rate, evoked response, temporal history, or an equally capable model of the raw signal. The proposed solution is to estimate those relations, test whether they improve held-out prediction, destroy them selectively with matched nulls, and reserve causal language for a later intervention. Only after this problem, operation, and test are stated do we give the proposal its name. Tomographic Coherence-Based Neural Rendering (TCNR) proposes that distributed, partial measurements can be stitched into a task-relevant state estimate partly through structured phase relations. This paper has two explicitly firewalled layers. Part I turns one bounded neural claim into a prospective protocol using the public Brochier et al. instructed-delay reach-to-grasp recordings from two macaques. It asks whether beta-band spatial phase organization during the delay improves held-out prediction of the cue-instructed grip condition under the declared representation classes, compared with evoked activity, spike-rate summaries, amplitude, power, scalar coherence, non-oscillatory temporal dynamics, an unengineered 1-kHz LFP comparator, and an equal-compute nonlinear LFP comparator. One animal is used for development; all feature definitions, model-selection rules, nulls, and decision thresholds are then locked before the second animal is evaluated. The primary estimand is the cross-fitted binary log-loss improvement of the phase-structured model over the strongest locked comparator, with 0.02 bits/trial as the smallest effect of interest. Phase scrambling, block-preserving label permutation, circular shifting, coordinate permutation, evoked-response controls, and a reference/phase-estimator sensitivity set test whether any gain is explained by power, common input, geometry, filtering, or the declared model class. In the source task, the fixed visual cue specifies PG versus SG; cue identity and grip instruction are therefore perfectly confounded. This dataset cannot identify a maintained motor plan independently of cue identity. This is a protocol paper. It reports no dataset result and does not test consciousness, causal necessity, or literal tomography. Part II develops a conceptual bridge from relational differences to higher-order patterns, prediction, action, and feedback. It distinguishes statistical dependence, neural oscillatory coherence, computational integration, and quantum coherence; places TCNR beside current consciousness-science comparators; and asks what a physical implementation would need to show before cross-scale language becomes more than analogy. The physics discussion is conditional and local: it does not infer quantum coherence from neural synchrony, identify spacetime with a computer, or treat coherence as sufficient for mind. It also distinguishes a physical difference from a difference available through a particular measurement channel, treats hidden-source recovery as an identifiability problem, and separates classical synchronization from quantum entanglement. The combined endpoint is a theory-and-methods paper. It reports no dataset result, consciousness measurement, cross-substrate validation, or physical unification.