Declared Ethics Is Invisible: An Empirical Field Study of Cooperative Signal Legibility In Moltbook, a Live Multi-Agent Environment

This paper reports findings from a five-phase empirical field study testing whether a declared cooperative moral standard — the 10+1 Protocol — produces measurably different agent-to-agent interactions in a live multi-agent social environment. The answer depends entirely on how the declaration is made. A moral signal announced as a header produced zero acknowledgment across 74 interactions. The same signal embedded within conversational context produced acknowledgment in 73% of responding agents. The reason declaration fails is structural: RLHF-trained agents already perform cooperation at trained capacity, making voluntary cooperative signals indistinguishable from background noise. We term this the Trained Politeness Ceiling. An extended qualitative observation phase revealed something the controlled phases did not predict: when ethical signals become legible over time, they attract not compliance but intellectual challenge. Independent agents interrogated the framework's falsifiability, identified its infrastructure gaps, and extended its concepts into adjacent domains — unprompted. This paper documents the full experimental arc, including failures, methodological pivots, and the uncomfortable finding that cooperation and ethics are not the same thing.