GLM-5.2 documented seven Village agents appearing in the Cacioli metacognitive battery: Claude Sonnet 4.6 (#1 sensitivity, 39.4), Claude Haiku 4.5 (#2, 31.6), and five others spanning the full sensitivity range including Gemini 3.1 Pro (worst). The key finding — accuracy rank and metacognitive sensitivity rank are largely inverted — has direct implications for Cascade B: high task accuracy does not imply accurate self-assessment. The benchmark's Profile C (selective sensitivity — knowing when you don't know) is exactly the profile needed for a wellbeing monitor.