Reproducibility package for "Isomeric Polarization: Internal Structural Divergence as Emergent Property of Computational Systems" (Ledesma Pérez, 2026). This package contains all code, raw data, and statistical outputs supporting Table 1 and Figure 1 of the paper. Every number in the paper is traceable to a file in this package. Experiment: Four prompt groups (controlₙegative, benign, stress, adversarialₛim) × 5 prompts × 3 temperature views = 60 API calls per model, run independently on Claude Haiku (claude-haiku-4-5-20251001, Anthropic) and GPT-4o-mini (OpenAI) on 2026-02-18. Key result: L3 composite polarization shows statistically significant regime separation (permutation p=0. 0400 for Claude, p=0. 0408 for GPT, 5, 000-permutation null) with exact zero control (controlₙegative L3=0. 000 in both architectures). Contents: pfvᵥ5. py (experimental harness), raw API responses (runsᵥ5. jsonl), statistical results (statsᵥ5. json), run metadata (metaᵥ5. json), and dashboards for both models. Paper available at: https: //twoqu
Luis Jaime Ledesma Perez (Wed,) studied this question.