I think of this as the "iterated lying" and "accumulated models constitutes understanding" approach to science education. :)
Consider giving a briefing in the military, or to business management, or in a professional consult. Yes, you need to simplify. But distorting results, missing the point, saying things with little connection to reality, and being disorganized and incoherent, rises towards professional misconduct. Describing a foreign culture to generals, or tech to middle management, doesn't seem incomparable to describing the physical world to kids. Science education content currently gives a really wretched briefing on the physical world.
> student [...] demolishing parts of the simpler [model] that no longer fit [...] This goes on until you get to [...]
becoming a researcher
But that doesn't appear to be what's happening now.
Chemistry education research describes chemistry education content as incoherent, leaving both teachers and students deeply steeped in misconceptions.
But maybe in college? First-tier astronomy graduate students are often unable tell a 5-year old what color the Sun is, without getting it wrong. First-tier medical school graduate students seem to have little grasp of cell size. These misconceptions and gaps then impair other foundational understanding. It is hard to alter misconception ecologies. Think kudzu.
But maybe researchers? Active researchers chop back their own kudzu in their specific research area. And variously trim it in areas they teach. But the ramshackleness of people's understanding increases rapidly, even within their field, as you move away from their narrow research area, and off into their kudzu forest. Where everyone else lives.
> [paraphrased:] a balancing act of introducing and dismantling increasingly detailed and intricate models
What does "teach friction" mean? Does it mean teaching kindergarten kids what can make them slip and fall? And behavioral and engineering strategies for avoiding that? Sock nubbies need to be on the bottom? Or does it mean teaching them, years later, algebraic plug-and-chug of Arrhenius' law of large objects sliding on pig fat? And for the pig fat model, should they develop a feel for reasonable sliding numbers? Should they be able to judge how well unfamiliar situations match the model?
Imagine being in a 19th-century one-room schoolhouse with a book or two. You might aspire to teaching plug-and-chug, on models students won't use in the real world. Imagine being in a 2017 classroom. You have plug-and-chug. You might aspire to teaching numeracy and transferable knowledge (can be applied to unfamiliar problems), but it would be hard, given the constraints you face. Now imagine a 2037 classroom. The kids have had AR their entire lives. Hybrid computer-human systems have dropped the currently ghastly-high cost of pulling together insights from very large numbers of busy and expensive domain experts. So what could you aspire to?
Might we aspire to a hands-on deep-and-broad understanding of the physical world? It's straightforward to teach early primary students bits of foundational knowledge that graduate students should have, but often don't. It's currently too expensive to scale that. Even considering the positive feedback of getting things "right". But costs are declining. Maybe we'll hit a new regime, as bizarrely unfamiliar, as say expecting peasants to learn to read?