They’re handy for a niche variety of tests for sure. Or when coverage was sketchy to begin with and you need a quick and dirty regression test.

Come try our tap water in Finland, cleanest in the world.

https://waterdefense.org/water/tap/countries-with-the-cleanest-water/#1-finland

Yes that's one way. Better yet drop some of the unnecessary characters and reconcile cases where the pronunciation would be ambiguous. (ie. perhaps "twice" should be spelled "twais" to make the special pronunciation rule about the letters "i-c-e" unnecessary).

I agree for sure. It's just that it's happened in a highly distributed uncoordinated way (which is absolutely amazing in and of itself) and I wonder if some concerted effort could yield faster/better optimizations.

Why aren't we spending more to optimizing it?

I mean we certainly have and it evolves on its own.... but such a fundamental component deserves more directed attention. Also fascinating is the role that emojis have played recently, as a form of written language. They transcend spoken languages unlike alphabets, yet their meaning is far less precise.