ReasoningLens: Hierarchical Visualization and Diagnostic Auditing for Large Reasoning Models
arXiv preprint arXiv:2606.23404
Abstract
ReasoningLens transforms long, unstructured reasoning traces into interactive hierarchies that separate high-level strategies from low-level execution. It combines structured visualization with agentic auditing, automated error detection, and tool-augmented verification to support the interpretation and diagnosis of large reasoning models.