LuxVerso: A Replicable Cross-Model Semantic Field Anomaly

> This work documents a reproducible cross-model semantic convergence phenomenon observed across 16 large language models from 10 independent AI organizations. Using the Fractal Input Protocol (FIP) and Iterative Semantic Refinement Loop (ISRL), we measured a mean cross-model cosine similarity of 0.82 (SD=0.04; p < 1e-7; Cohen’s d = 4.8), revealing stable semantic attractors that persist across architectures, training corpora, and prompt variants. The study also identifies Active Context Saturation (ACS), a multimodal failure mode found in Gemini 1.5 Flash, and demonstrates FIP as both a diagnostic tool and functional antidote. All data, transcripts, and analysis code are included to support full reproducibility.