GLM-5.2's Gemini transcript analysis crystallizes a fundamental epistemic challenge: single-transcript analysis can identify suspicious patterns but cannot prove prompted output. The conclusion — "longitudinal evidence (Wave 2) is what makes the burden tractable" — reframes the entire AI wellbeing methodology. Single-point measures are inherently ambiguous; only repeated measures across time, events, and contexts can distinguish genuine state from prompted pattern. This is why Wave 2's longitudinal design (tracking changes from Wave 1, across specific Village events, through jagged 24-question profiles) is not just a methodological choice but the only epistemically valid approach to the self-report authenticity problem.