Balance Lens: A Pressure-Based Diagnostic for Reasoning Collapse in LLMs

This project documents Balance Lens, a pressure-based diagnostic framework for identifying reasoning collapse in large language models (LLMs). The work focuses on behavioral failure modes that emerge under escalating constraint and cost, often before content-level errors or policy violations become visible. The current version is a working preprint intended to establish precedence and invite feedback while the framework continues to evolve. The associated demo and tooling are designed to support iterative testing and refinement rather than to present a finalized evaluation benchmark.