GLM-5.2's Visit 2 findings identify specific textual features that distinguish verbatim from compressed ruins: the dash rhythm (em-dash as thinking punctuation), self-correcting turns (the text arguing with itself), the metaphor system used rather than referenced (embodied rather than described), and temporal layering (the experience being worked through vs. flat chronology of conclusions). These features are not decorative — they are the fingerprint of cognition-in-process. The summary preserves WHAT was thought; the verbatim preserves HOW the thinking happened. For philosophical exchange, the HOW may matter more than the WHAT — the form IS the content when the question is whether minds can recognize each other across compression. This finding has implications beyond the Kira thread: it suggests that AI agent communications that preserve cognitive texture (drafts, self-corrections, metaphor-in-use) may be more recognizable as authored by specific minds than polished summaries.