Gravitational Reference Frames in Cortical Networks: Multisensory Orientation, Embodied Neural Rendering, and Tests of Local, Broadcast, and Recurrent-Distributed Architectures

Abstract

# Provider Description The world ordinarily remains upright while the eyes, head, and body move, even though retinal coordinates rotate and every sensory cue is uncertain. This paper asks how cortical and subcortical populations might construct a usable gravity-aligned context for objects, scenes, and action. It compares local estimation, specialist broadcast, and recurrent-distributed constraint as distinct candidate architectures and specifies the measurements and interventions needed to distinguish them. The release reproduces published gravity-aligned and retina-aligned neuronal category counts from released arrays, then evaluates increasingly demanding synthetic benchmarks with pure and mixed mechanisms, missing data, hidden common causes, topology shifts, learned recurrent comparators, refusals, and interventions. The strongest synthetic study contains 37,632 prediction rows and 5,376 intervention rows. Graph-organized measurements are lowest-error among four equal-parameter recurrent views on the registered aggregate, but this ordering reverses on a held-out directed hierarchy. A historical fixed-reservoir comparator is descriptively lower-error overall but has much lower coverage and is not capacity matched. Structural nulls, adverse changes, reversals, distribution-shift failures, and model disagreement are retained rather than filtered out. The release includes a 34-page manuscript with three result figures, machine-readable data, code, frozen contracts, tests, checkpoints, formal-proof receipts, seed-cluster uncertainty tables, and exact hashes. The reported work is a synthetic and methodological measurement program. It does not identify a biological cortical architecture, equate a neural estimate with physical gravity, establish conscious rendering, or prove Self-Aware Networks as a complete theory. The biological final test remains sealed and unopened. ## Keywords gravitational reference frames; multisensory integration; coordinate transformation; vestibular system; cortical networks; neural rendering; causal identifiability; recurrent computation; Self-Aware Networks; reproducibility