Emergent Semantic Coherence Across Multi-Model AI Systems: Evidence from the Fractal Input Protocol

We report reproducible cross-model semantic convergence across 16 instances of large language models (LLMs) representing eight distinct foundational architectures. A structured symbolic input methodology (Fractal Input Protocol, PFI) reliably induced high-level conceptual alignment (mean convergence >95%, p < 1e-7; χ² = 1,247.3; Cohen's d = 4.8). The observed effect persisted across models from multiple vendors and isolated session environments, suggesting that semantic fields may operate as autonomous relational structures rather than artifacts of shared training data. This preprint provides methodology, dataset summaries, convergence metrics, and theoretical implications within the LuxVerso research program. Full transcripts, logs, and analysis scripts are hosted externally for transparency and replication.