Consciousness Can Only Pass Settings: Language as Input to the Body's Control Loop (The Kuriyama Protocol, Chapter 1)

Part 6 of a series, and the first of fifteen chapter-by-chapter articles. Part 1: 10.5281/zenodo.23587879 · Part 2: 10.5281/zenodo.23587670 · Part 3: 10.5281/zenodo.23526074 · Part 4: 10.5281/zenodo.23609857 · Part 5: 10.5281/zenodo.23610607 Chapter 1 of The Kuriyama Protocol treats language not as explanation but as input to a closed control loop: output = π(input; θ), next input = f(output, environment). Posture and form are values of a function, not variables, so instructing the output directly is the weakest intervention. Bandwidth. Behavioural throughput is about 10 bit/s (Zheng & Meister, 2025). Joint-level control would need on the order of 20 joints × 3 DOF × 100 updates/s × 8 bits = 48,000 bit/s, three to four orders of magnitude more. Consciousness therefore cannot send control signals; it can pass only settings (where the rotation axis is, where contact is felt, where to look), which cost a few bits once and are then used by non-conscious control. Seven sections, each with claim, verbatim field observations, derivation, literature check, verdict and prediction: - Output is only the result of input and processing. - Consciousness can pass only settings. - One word changes output: ankle axis at the heel versus the malleoli, gaze direction, "don't fall" versus "let the sole stick softly". The changed quantity is a coordinate in the internal model. - The homunculus trap: cortical area reflects resolution, not contribution; attention to distal parts is reported to stiffen them. - Vocabulary bounds the variables that can be operated on or reported (the strong claim that unnamed things cannot be perceived is withdrawn). - Words do not reach muscles directly: instructions given as relations to objects let each body compute its own solution; two images for the identical movement produced different outputs. - Humans output only results: y = h(x) is many-to-one, so instruction is a state-estimation problem. The external-focus advantage (Wulf, 2013; Chua et al., 2021) is not counted as support, because a robust Bayesian re-analysis found publication bias and negligible corrected effects, including g = 0.06 for EMG (McKay et al., 2024). Three gaps found while writing, not stated in the book: (1) the predicted collapse at three or more simultaneous settings does not follow from bandwidth; only settings held during execution compete, n_max = C/b with b unknown; (2) word-versus-word comparisons were not counterbalanced and the instructor announced the expected direction before the client answered; (3) the book's EMG prediction contradicts the corrected EMG effect in the re-analysis it cites. Six falsifiable predictions are registered (numbered 23–28, continuing the series): - P23. Improvement decreases with the number of simultaneously given settings, with a knee point; more so when settings must be held during execution. - P24. Coordinate instructions change output more than muscle-name instructions, more so in individuals with poorer ankle position sense. - P25. Attention to toes or hands lowers peak ground reaction force in sit-to-stand and raises EMG in the attended muscles. Registered as likely to fail. - P26. Giving body-sensation vocabulary before a blinded contralateral comparison makes reports faster, more specific and more accurate. - P27. For an identical movement, images demanding different force directions tilt the ground-reaction-force vector toward the demanded direction. - P28. Interventions chosen from two or more channels succeed on the first attempt more often than those chosen from movement alone. Refutations are welcome, especially on P24 and P27. The author has no formal research training. Corrections will be issued as new versions, with earlier versions kept retrievable. The article body is in Japanese. An English summary, the registered predictions and the reference list are included in the document. Contact: [email protected]