UTS Team Places Second in Psychological Defense Mechanism Classification

A group from the University of Technology Sydney (UTS) secured the second position (F1 0.406) out of 64 participants in the PsyDefDetect challenge, which focuses on identifying psychological defense mechanisms in emotional support conversations via the Defense Mechanism Rating Scales (DMRS). Their significant finding was that defense mechanisms are characterized by what is lacking—such as absent affect, blocked cognition, and denied reality—represented on an affect-cognition integration spectrum through prompt-level clinical rules, resulting in the highest single improvement (+11.4pp F1). The system employs a multi-phase deliberative council of Gemini 2.5 agents for class-specific evidence evaluation, achieving F1 0.382 without fine-tuning (a top-5 finish). Nonetheless, the council often misjudged minority classes, with 59-80% of stable minority predictions being incorrect due to a consistent "L7 attractor" bias towards the majority class, which was addressed with a targeted override ensemble.