Very rambly post incoming.
The question of where does "learning" stop and "applying" begin in games is very interesting because it's like a central point that branches off into many interesting discussions.
The learning doesn't truly stop once you've gotten some pre-requisite pieces of information because applying it generates new information which then becomes part of your information pool. Additionally, games are logical systems and the optimal solutions are defined in the code, so they rely entirely on obfuscating optimal solutions from the player.
So how does the distinction make sense? I think this can be somewhat resolved by looking at learning in games as a density + interconnectivity graph - the more you learn the more dense & interconnected the elements you have to consider become. When you pass an arbitrary point of the spectrum, you enter "applying" territory, where the elements are so interdependent and dense that figuring out the optimal one requires you do very fuzzy internal calculations and weighing many different things.
So you learn when you engage with the least coherent parts of the games, and apply when you engage with the most coherent parts of the games. In my opinion the magic of games lies at the parts where they cohere the most.
All this abstract crap was leading to a simple thing - I recently realized that despite playing them a fair bit, I fundamentally don't like the focus of most new 3D physics platformers that much. Specifically because their learning phase (filtering through least coherent elements) sticks around and interrupts the applying phase (filtering out the most coherent elements), the latter being what I'm after when I play these games.
The games are subtly different in design philosophy to racing games, to the point where I'm tempted to call them games about optimized exploration. Racing games (and some older physics platformers), while certainly not being free of massive skips and complex tracks/levels, tend to make an honest attempt at rushing you through the learning phase. They do it by showing more of the level at once (or the whole level), having maps, keeping things linear, minimizing the presence of skips, having short repeating levels (cuz of laps), making each level a very distinct chunk of a game through visuals/shape/theme, etc.
Newer games seemingly fully commit to pleasing speedrunners, so the levels tend to be a lot more dense, with more routes, much more complex geometry that invites abuse and skips, and a general exploration based approach. What this does is create a lengthy period of a game where youre collecting the puzzle pieces before you can start putting the puzzle together, and constant interruptions where your organic step-by-step routing and improvement gets invalidated because you noticed a skip. And this cuts against the core appeal these games have for me - godlike, smooth, intuitive learning of the most cohesive parts of the game.
Some games like Umihara Kawase are exceptions to me because, despite having a lot of skips, the learning curve is more zigzaggy in general so the skips don't distract from it.
But maybe the salt from learning a lot of all-or-nothing skips I'm not consistent enough at accrued and I'm just being a baby. Only time will tell.
Penny's Big Breakaway is still my GOTY tho