If you consider how knowledge decays, your strategy makes more sense. You get to something like Pareto proficiency in a subject and can then ‘reactivate’ that knowledge more easily in the future, should it be required.
You are also of course aware of that knowledge in the first place, which may be even more important. In the timeless Rumsfeldian: you have pushed back a bit at the unknown unknowns.
The alternative path of deeply learning something you may never apply (e.g. what happens to many PhDs) seems inherently less desirable, in my opinion.
I'm at about the same place with the TLA+ videos. I finished them a year ago and didn't feel like I could start a project yet, I more feel like I should watch them a second time. I have an actor system project on the side that I've felt tempted to model using TLA+.
Right now I got my hands on Tufte's "The Visual Display of Quantitative Information" as per some HNer recommendation (thank you!).
Yeah, I lose interest quickly, eh. There's so much cool things to learn that in the end I learn nothing well. Bummer.
edit: grammar, spelling