ZLT-50L-B The Pharisee in the Machine: CM-002 Failure Confirmed in LLM Inference via Distributed State Tracking; Cross-Model Comparison Reveals Three ZLT Regimes Under K=0

Read full text Summarise with VIRAT
Authors: DANNY YUBI DAGOGLIANO
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Large language models in inference have no backpropagation. Their weights are fixed. Context accumulates — and nothing corrects it. Zero Leap Theory predicts that this structural condition (K=0) produces a specific, observable failure mode when context length exceeds a critical threshold L*: CM-002 Pharisee Failure — the system applies the correct reasoning policy over a corrupted internal state, producing output that looks like thinking and is empirically wrong. This paper reports the experimental confirmation of that prediction. The central finding (Claude Sonnet 4.6, Register Machine task): At L=4,000 tokens of irrelevant context with ops=4 fixed (same as baseline), a McNemar paired test (N=40) yields b=7, c=0, chi2=5.143, p

Fields of Study & Subjects

Topic Modeling Natural Language Processing Techniques Adversarial Robustness in Machine Learning

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN
Document type
Preprint