But they wanted a lot more than that. Engines like Phaser work hard to take care of browser quirks for things like PointerEvents vs. TouchEvents vs. MouseEvents or supporting mobile devices. Sure, it seems simple at first until you run into those kinds of problems and reinvent the wheel... learning an engine isn't terribly complicated but I understand the sentiment for a one-time project. It just seems like they did so much other planning but didn't want to plan the UI implementation to the same degree?

Like you said, it was a one-time project that wasn't incredibly complex and just needed a one time deployment. And they had UI guys who generally knew what they wanted to and how to do it. I think it would probably have taken them more time to research suitable engines they could rely on than to build the functionality themselves (or use whatever libraries they were already very familiar with.) All engines/libraries end up having quirks that you really only learn through experience.

Also they were kind of time constrained. WebGL would have been definitely more performant but it has all sorts of hardware issues at times in different devices. I've done plenty of WebGL over the years and seen random GPU crashes where you have to restart the entire machine to unlock your self out.

The canvas trick with typed arrays is brilliant. Using requestAnimationFrame is what any front end dev who knows perf would do. Its like the front end version of cdn with 1 second time out trick.

Also using a layer of library whose code you don't understand in a perf critical application is quite risky. Its better to stick closer to the native browser APIs which you are familiar with. We once had to throw away a library and rewrite code from scratch because their assumptions failed when pushed to the limits. The rewritten code was 100x smaller in size and 10x more performant.

I would have loved to work on something like this but it sucks reddit doesn't have any presence in Seattle.