Structuring Reasoning for Interpretability and Control

The reasoning process is opaque.

Large Reasoning Models generate long, unstructured traces — and we can't see inside them.

LRMs solve complex problems by allocating substantial computation before answering. Rather than producing a response directly, they generate a long reasoning trace and condition the final answer on it. These models have achieved strong performance gains in mathematical reasoning, code generation and multi-step inference.

Yet they introduce a new and largely unaddressed challenge: their reasoning process is unstructured, and thus neither interpretable nor controllable. Models output monolithic sequences of thousands of tokens, do not explicitly segment or annotate reasoning steps during generation, and their internal representations do not expose clean boundaries between reasoning stages.