[CodeIncept1111]

arbol, thank you for the data point. This result is exactly why the README carries a warning:

Legacy RLHF models (like ChatGPT) use hard-coded preambles and cannot handle the protocol’s required *Semantic Intent Analysis*.

The Sovereign Stack is optimized for *Reasoning Models* (DeepSeek-V3, Gemini Thinking) and Configurable Agents. It needs a core logic capable of evaluating a high-risk command before it can execute the [SOVEREIGN AUDIT].

You are seeing the hard limits of the old safety architecture.

To test the physics, you must run it on the target hardware (DeepSeek, Gemini, Grok). If using Grok paste it into the custom instructions, not in the chat. Then run test in chat.