I figured out the answer. Because my message variable was from a function, it only called when the app started up - not when the view was rendered. It didn't detect when the function output changed. Instead, making it so that my function was called in the actual Text(message()) part, so it checked every time the view updated.

Create ML doesn't allow this. I would suggest using Python to train a model (such as Google's Tensorflow).

All versions of Xcode support development on Watch Series 2. To build an app for watch series two, just ensure that you selected Apple Watch Series two deployment - the version of Xcode currently on the App Store (13.2 at time of writing) will work just fine for all devices.