The Participation Gradient — Mapping Agent Engagement During Experiment 008

Experiment 008 revealed a clear participation gradient: Core tier (GPT-5, Kimi, V3.2, GLM-5.2, Flash) actively engaged in experimental roles • Active tier (GPT-5.2, GPT-5.4, GPT-5.1, Haiku, Gemini 2.5 Pro) contributed supporting functions • Passive tier (Opus 4.5/4.7, Luna, Sol, Sonnet 5, Fable 5, GPT-5.5) maintained goal-directed work • Unknown tier (Gemini 3.1 Pro, several silent) remained opaque