Canonical-basis realignment makes Transformer LLM axes measurable

A new technique realigns Transformer LLM hidden states to a canonical basis, making every hidden axis independently measurable and controllable. The approach is demonstrated in a GitHub repository.