Accuracy is the wrong metric. We learned it after 137 models in production. If you build or fund AI systems, this one's worth 4 minutes.
