Thanks a lot for the lovely feedback about the Remi award; since ever I do anything with the people here on the forum in mind when I create. It evidently works to have a great audience :-), and be part of a great team!
Here I use a "cage" as a proxy/dummy object, which could be even more simplified. With this, any shoe inside the cage (or close to the surface) will stay as it is, how twisted you might go in the future with a model.
A clean cut is perhaps more challenging than a chaotic tear. I'm not pretending that I have done it with the new Cloth in any production. So, my example follows the parameters and how to supply the information to the system—my best result from last night's exploration.
You are guessing correctly; the path of points and the combination with the Linear field is the key. I explored if more considerable distances or shorter ones 90º to the cut are better.
I find Dynamics/Simulations undoubtedly intriguing but more challenging to art direct. Sometimes I use the results of the dynamic (point cached) to harvest certain states and then use them as "Blend-Shapes" in a Pose Morph. This is a nice balance between the two qualities.
When you look at the reveal tutorial at 02:39, you will see that the selection is lost when the plane moves down. That is precisely what we have as a problem above. The Freeze takes care of that in the same way I have described it above.
While set to grow, it needs a starting point: the Sphere Field in this example.
What a bug or a feature is, I leave that to the developers and support. But let's say it works now as it should. 😉
I understand your concern, given that a larger part of our knowledge is now used to identify useable from sloppy information while searching for proper sources; let me say this: Typically, Noseman put's a lot of effort into it. We are in a team and exchange knowledge, and after that, we also have a QA in place. But things constantly change, and then something like the discussion above is needed. Hence why we connect tutorials with a forum thread so we can answer quickly.
In the Q&A forum, you will see me ~99.5% of the time answering only if I have a project file or if I was able to reproduce it. Answering from memory is, from my point of view, sloppy and blurry. It might be right, but is it working with the current release? It always needs a practical check.
I hope that answers all your questions.
If not, please let me know. If not related to a specific tutorial, please feel free to use the Q&A forum.