Interactively inspect how hidden‑state patching (Patchscopes) reveals a word's detokenised representation across model layers.