There's a couple of approaches and many will suggest the "small games" route.
I disagree.
I believe that you should make a list of ten(or more) ideas you have, pick the one you like the most, regardless of scope, and try and do that.
You can 'over scope' like this, especially if you add the hard mode modifiers like online play, 3D, Custom Engine dev(at least at the start), and a couple others, but honestly, it's better to try and fail at something you like rather then zero in on picking something small. What's the risk? You try to make the next Call of Duty for a month and realize that netcode and realistic military texturing is too much of a lift? Most people try to optimize at the start of the process, but you should optimize constantly instead, so starting point is irrelevant.
>>717261770
The one thing I saw from Redot's roadmap that stuck with me are supports for Structs within GDscript, which I guess would be analogous to Dataclasses within Python. I like the idea of doing that over creating custom resources for my data.
Regardless, I wouldn't consider a fork that doesn't have a year or so of development behind it OR isn't a result of some community disruption like what happened with PyGame.