One Successful Prediction, Not Four: Method, or Why These Claims Can Be Made (The Kuriyama Protocol, Chapter 0.5)

Part 20 of a series; the fifteenth and last of fifteen chapter-by-chapter articles. Part 1: 10.5281/zenodo.23587879 · Part 2: 10.5281/zenodo.23587670 · Part 3 (v2): 10.5281/zenodo.23723510 · Part 4: 10.5281/zenodo.23609857 · Part 5: 10.5281/zenodo.23610607 · Part 6: 10.5281/zenodo.23682055 · Part 7: 10.5281/zenodo.23684537 · Part 8: 10.5281/zenodo.23684996 · Part 9: 10.5281/zenodo.23685960 · Part 10: 10.5281/zenodo.23686289 · Part 11: 10.5281/zenodo.23720048 · Part 12: 10.5281/zenodo.23723725 · Part 13: 10.5281/zenodo.23834848 · Part 14: 10.5281/zenodo.23854475 · Part 15 (v2): 10.5281/zenodo.23886603 · Part 16: 10.5281/zenodo.23859894 · Part 17: 10.5281/zenodo.23874674 · Part 18: 10.5281/zenodo.23886834 · Part 19: 10.5281/zenodo.23903352 Chapter 0.5 of The Kuriyama Protocol sets out the book's method: six procedures for within-person causal inference without statistics (contralateral self-comparison, stopping rule, forward-only search, compensation-triggered debugging, deception-cost ordering, author-as-subject), AI as a falsification device that lowers the cost of doubt, eight recorded rejections, and four cases in which theory predicted and data later agreed. Five gaps found while writing: - Of the four "successful predictions", only one survives the series: arm swing driven by vertical oscillation failed by calculation (Part 14); only the walk–run transition at Fr ≈ 0.5 holds (Part 11); the compensation–minimal-intervention match is inverted, since the principle treats task-achieving redundancy as optimal (Todorov & Jordan, 2002); and "phase cannot be given from outside" was rewritten (Part 18). - Contralateral comparison assumes an unchanged control side, but unilateral stretching increases range of motion in non-stretched joints (Behm et al., 2021). - Resistance to faking is not accuracy: unintentional noise (light for pupils) was omitted, and nonverbal cues such as self-touch are weak indicators (DePaulo et al., 2003). - Three earlier corrections had not reached the method chapter. - The rejection record stopped at eight, and its summary ("reasons fell, never claims") no longer holds. Five falsifiable predictions are registered (numbered 92–96, continuing the series; 96 in total): - P92. Unilateral treatment widens range of motion on the untreated side relative to a no-treatment day. - P93. Pupil-based arousal agrees poorly with heart rate under uncontrolled lighting and better under matched lighting. - P94. Correcting task-achieving compensation leaves immediate performance unchanged but reduces load complaints within weeks. - P95. Smaller ankle dorsiflexion predicts more knee valgus in squatting; increasing dorsiflexion reduces valgus without knee instructions. - P96. One-at-a-time interventions yield smaller immediate gains but higher next-session replication than simultaneous ones. Refutations are welcome. 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]