At Exponential, about 20% of the risk-scoring is automated. How much can you just automate away and then write code to analyze? We're going to do a lot of research, which ends up being something more than 40 hours of research, as Greg mentioned, just for one. In protocols again, we are going to look at the Lindy effect, like how long had the protocol been around saying that a protocol that's really new, it probably a riskier than a protocol that has been battle tested.