Happy to answer questions.

This is early, but the goal is simple:

Make AI execution verifiable, not just outputs.