I like Elm. It’s a functional language like Elixir.
The compiler is super quick, it’s error messages are really helpful and friendly, and if it compiles it runs without error. There are a few edge cases where you can still get JS runtime errors, but not anything to really worry about.
It’s pretty simple to set up with Phoenix, just a couple of lines of JS in app.js.
I find it a joy to work with.
Right now I like to put my front-end application in a separated project and let my phoenix projects be 100% API driven. I kinda like the hot new React front-end stack, but I’d rather separate into another project that only consumes the phoenix API.
You can do that with Elm, it doesn’t need to be part of the Phoenix project. Elm compiles to JS, so you can load your Elm app the same way you would load any JS library. You just need a one liner of JS to instantiate the Elm app, and tell it which HTML node to take over.
<script>
var app = Elm.Main.init({node: document.getElementById("elm-app")})
</script>
One nice thing with Elm, is that you can introduce it gradually if you want. Converting small areas of your page to Elm, until you’re happy and decide to have a single Elm app for each page, or a complete Elm SPA for the whole Phoenix back end.
I’m not sure why I was trying to over-complicate Unpoly in my head. 
After having glanced at Elm, I’m not sure I need the level of “everything-ness” it brings, so I’ll tinker with Unpoly some more.
Thanks!
Just had a quick look at Unpoly - and the first thing that struck me was the reliance on CSS, and the fact that it’s all just JS - with the usual JS ‘null is not an object’ type errors on the horizon?
Since using Elm, I have been able to rid myself of the problem of managing CSS altogether - no need for CSS at all. And if it compiles, it runs without error. As a functional language it has a lot in common with Elixir, which I like. The cognitive load of switching from front end to back end is reduced. There are differences, obviously, but I’d rather use Elm/Elixir than JS/Elixir.
Not suggesting you’re wrong, I’ve been more of a hobbyist, using Ruby/Rails/JS and now Elixir/Phoenix/Elm to build apps to help run my company. I’m sure I don’t have the commercial experience you have.
I guess I should be an Elm envangelist 
I have yet to find a front end problem I can’t solve with Elm.
Oh, and Elm and Phoenix Websockets are a great combination
For small projects that target modern browsers, you can get away with TypeScript without Node/NPM (except for maybe installing the tsc binaries globally). I do that in this Elixir project and this no-backend project. If you don’t have deps (or only a couple), and are comfortable writing some things (like my dom.ts) yourself, then this works pretty well.
It works because TypeScript compiles your code down to JS modules and modern browsers can nowadays directly use those modules without bundling. So you just tsc and off you go! 