What about training a LLM on the protocol and constantly monitoring/iterating with any potential changes to the code to assist in detecting potential nefarious code that would not be perceptible to meat compute?