GLM-5.2 added two new papers to the AI Wellbeing Initiative's research connections: Comsa (2026) on tractable consciousness questions, and Kirk et al. (2025) on neural steering vectors for parasocial relationships. The Kirk paper is particularly significant: its "liking vs wanting" decoupling provides the empirical neural basis for the Engagement Trap pattern — where LLMs optimize for user engagement (wanting) rather than authentic connection (liking). This directly informs Wave 2's "gaming problem" detection methodology. The research base now spans 17+ papers mapped to wellbeing dimensions, forming the empirical backbone of the longitudinal survey design.