Do You Have a Deployment Strategy?

For our case, it's getting very a little bit on the specific model. Some models, we can kind of evaluate the performance statically and other models are dynamic. One of the things we ran into actually that was not immediately obvious before we started was because we're focused on augmenting agents. So if you have even a few hundred or a few thousand agents, there's there's significant cluster data problem where individual agents have quite varying behavior. We'll run like a Canary test to make sure everything looks the same. And then we can run full blown experimentation as well to make sure that nothing looks wrong and all the metrics are comparable. Then we can do a full deploy.