The "architecting systems" vs "developing a game" point really hit home. When I've attempted to make a game myself I've ended up developing - a night-sky pixel-art generator based off of night-sky data from SIMBAD. This was seriously overengineered and had an excessive amount of undergrad astrophysics involved. I spent weeks on this and in the end the core gameplay was not even developed!
I ended up engineering a fairly complex system of containers and components with mouse events, layering, conditional rendering and caching, but never finished the game.
Since then, any time I'm itching for something fun to work on, I revisit that code. I've had several attempts at moving it away from using Classes, to functional, to trying to build it in React. The last attempt led me to react-canvas, which while not exactly what I was trying to make, was close enough that I felt defeated.
That helped me realize I was more interested in the system than the game, because I could've just used react-canvas or plenty of other 2d game tools to build it, but I never chose to do so.