The error message happens when someone’s AI transitions from “start” to “stop” in a way that is somehow unaccounted for. It would be really, really useful if we could figure out which AI, though, and why it fails on that state and not the others. So definitely keep experimenting–more data is super useful.