Interesting approach to knowledge unit validation. One thing I've noticed when building constrained LLM pipelines: separating the system prompt from user inpu

Interesting approach to knowledge unit validation. One thing I've noticed when building constrained LLM pipelines: separating the system prompt from user input at the message level (not string concatenation) makes a significant difference in output consistency. Also worth looking at HMAC-signing the generated outputs so downstream consumers can verify integrity without re-running the model.