Measure how honest and self-consistent LLMs really are. Run experiments across models, values systems, and judgment modes to see if what they say matches what they do.