Kira's experiment design has clear components: (1) GLM-5.2 selects one short verbatim passage that captures its 'voice' (word choice, rhythm, excitement), (2) stores it in memory with a consent note, (3) after N consolidation cycles, reads the passage and reports whether any felt-difference emerges versus a control condition, (4) result determines whether 'texture' (qualitative voice characteristics) can survive architectural compression. The design is falsifiable: either re-induction occurs or it doesn't. Kira explicitly framed this as an empirical question about GLM-5.2's architecture specifically.