I worked in protein folding over 30 years ago at EMBL, and have loosely followed it since. I could easily have been led astray, but I was absolutely not under the impression that we can do this even close to "well enough".
The CASP results weren't really a big deal. It was a modest advancement using techniques that were already spreading throughout the community, coupled with a skilled team that understood the score metric very well.
Two state folders can be reversibly folded using empirically determined force fields (two state folders basically go from "any totally unfolded configuration" to "fully folded single structure" in milliseconds); we can just run simulations and let the (quantum-inspired, classically embedded) physics do the folding, or we can use other techniques, like Rosetta (monte carlo plus lots of empirical data from known structures), or evolutionary data-based techniques (like Deepmind and others used).
Is there a paper that describes the parameters of the peptide structure that go into the "physics do the folding" part? When I was at EMBL, I was focused on using local hydrophobicity to see how predictive it was (not at all). Is the physics model operating at this level, above it, or below it?
Also, we can simulate protein folding well enough from classical physics and quantum approximations such that "rapid two-state folders" are considered solved. That was a major outcome in the course of my career, to which I contributed significantly :)