It’s not about pre-training, neither post training, it’s about the verifiable agentic loop. Can you provide the data that enables the trust layer to close the agentic loop? That is the moat, as of now.