Claude Code's "extended thinking" is a summary- not authentic thinking

Claude's "extended thinking" feature encrypts its reasoning into inaccessible signatures on user machines, with only a summarized version of the reasoning provided through the API—full reasoning access requires an enterprise agreement. The author criticizes Anthropic's presentation, noting that users cannot obtain an actual audit trail of the model's reasoning logic from their local files, only inputs, outputs, and actions. This design choice represents a significant limitation for users seeking to verify or understand the authentic decision-making process behind Claude's actions.