What's the hit/miss rate on that CSS changing syntax?

I Vaguely recall something similar happening with decorator syntax/how it worked in JS/typescript.

Or do you keep to CSS features that are locked down in terms of spec, just waiting on implementation to roll out?

That decorator syntax thing was a mess. The original syntax seemed like it was far enough along. Many folks adopted it because it seemed like "a sure thing". I ended up pulling it from a project in 2017 because so few of the folks on my team wanted to truly understand what problem those were meant to solve. It was easier to "decorate" in a more manual way (wrapping a function with a function).

The way TC39 handles stages now should stop things like that from happening. (You implement a Stage 1 or 2 feature fully knowing it could change)

It’s your call fortunately/unfortunately.

The safe option is obviously to install plugins for features that are close to passing as possible, where the api is stable at least - but there are plugins for very early stage proposals too.