UFAL/UBM v0.1: The Muñoz Number (N_M) — A Stability Margin and Reference Benchmark for Long-Context Reliability

This deposit releases UFAL/UBM v0.1, a reference package accompanying the technical report "The Munoz Number (N_M): A Universal Stability Metric for AI and Physical Systems". The release introduces the Munoz Number, defined as: N_M = -ln(F/C) as a dimensionless stability margin for adaptive predictive systems under sustained informational drive, and specifies UBM v0.1, a benchmark-style protocol for reproducible long-context reliability reporting. WHAT'S INCLUDED The full PDF technical report (UFAL/UBM v0.1 spec). knc_reference_refactored_ascii.py: KNC (Knee by Maximum Curvature) reference implementation with quantile binning, isotonic regression, geometric normalization, and bootstrap CI95. experiment_dual_refactored_ascii.py: a toy dual demonstration (viscous Burgers gradient-stress proxy vs. toy Transformer directional OD-Fisher proxy) illustrating UFAL's dynamic analogy at the level of stability operators. README.md with reproduction instructions. Minimal *.example.jsonl files documenting the expected log formats. SCOPE / LIMITATIONS The dual experiment is a toy illustration and is not a certification result. UBM v0.1 compliance requires laboratory-generated inference logs (token-level drive/conductance measures plus case-level failure labels). This release provides the specification and reference implementation to enable independent replication and standardized reporting. Related records (Zenodo): PPLM v2 (Principio de Propagación Limitada de Muñoz): 10.5281/zenodo.18326232 USL (Universal Stability Principle): 10.5281/zenodo.18340330 Conductance–hallucination empirical audit: 10.5281/zenodo.18615984 Structural Complexity Audit Suite (CLTM): 10.5281/zenodo.18637923 (Metadata update: cross-references added; files/content unchanged.)