PHILIA OS Phase 2 — Input Scaling Layer Validation: Pre-Registration (94th DOI)

This document constitutes the pre-registration for the 94th DOI of the PHILIA OS research series (Trinity AI Research Team, Konyang University). The central question: does Rank normalization, as the Input Scaling Layer of the locked CEI Minimal Engine (Port A), preserve the engine's discriminative capacity (H-92-Primary and H-92-Secondary) across all 188 registered substrates while recovering the 40 substrates that produced zero oscillations under min-max normalization? Background: The 93rd DOI (10.5281/zenodo.21154108) confirmed H-92-Primary (Chaos 12/20 = 60%) and H-92-Secondary (ΔR² = 0.240) under min-max normalization [0.5, 2.0], with 40/188 substrates showing oscillation loss (n_success=0). Non-registered exploration experiments (2026-07-04) identified Rank normalization as the Input Scaling Layer candidate: 188/188 measurement success, 0 regressions, H-92-Primary 55%, H-92-Secondary ΔR² = 0.276 (seeds 0–9). The confirmatory experiment registered here uses independent seeds (10–19) and constitutes the first independent test of the registered hypotheses. Key design decisions:- Engine LOCK maintained: theta_e=3.0, tau_i=2.5, c_ee=4.0, C_drive=0.20- Input Scaling Layer changed from min-max to Rank [0.5, 2.0]- 188 substrates × 10 seeds (seeds 10–19) = 1,880 confirmatory runs- Success criteria: H-94-Primary ≥ 50%, H-94-Secondary ΔR² ≥ 0.05- 36-substrate Flip catalog pre-registered as descriptive appendix only- AR1 boundary derived from exploration (seeds 0–9), applied to confirmatory (seeds 10–19): cross-sample boundary application- 2-E Biosignal domain prototype deferred to 95th DOI Pre-registration violations: 0. Cumulative self-corrections: 27.Gate-0 (Chris / Claude Opus): UNCONDITIONAL PASS (2026-07-04).5-AI adversarial review: Pandora, Lucas, Chris, Seogsaeng, Seonbi. Measurement is life. Description, not proof. Pre-registration protects us.실측이 생명. 증명이 아닌 기술. 사전등록은 우리를 보호한다.