Without having tried this, I'd say the problem with a predictive algorithm is that it is (ironically) impossible for the user to predict what will happen.
So after switching, they will need a short moment to reorient: understand where they were taken, check if it matches where they wanted to go, and then either switch again or stop the switching process to resume work. In UX design, it's better if you can complete a longer process without having to halt and reorient many times in the process (like opening a menu that was hidden and wait for a loading animation to complete, until you can actually read the menu items are).
If it's impossible to keep a mental model of where you are in the system, and how you can move to another specific window, then actually EVERY window switch requires much more effort and conscious thought.
I think windowing systems, virtual desktops, spotlights, stage managers, exposés, mission controls, are all too complicated... I don't know what the solution is, and I think it's great that people are working on novel solutions. But I do know I want to easily switch between 2-4 windows without the order randomly changing.
I have a very similar frustration with the complexity here, and found that scrolling tiled window managers (like PaperWM, Niri, etc.) might actually be the answer. All your windows are in a line, press the shortcut until you hit focus the one you want. Reorder with a shortcut or with a mouse.
The main problem with them is system support, they are buggy when tacked on top of a desktop OS (PaperWM), or require a pretty finicky custom setup (Niri).
This! I actually kinda miss the Win7 enhanced switcher (I actually kinda miss the whole Win7 Aero desktop, but I digress) ... Win-Tab bringing up a scrollable stack of the actual windows you have open and being able to either directly select the one you want or just keep Win-Tabbing through them. The Win10 tiled view is similar, but lacks the charm somehow...
I completely agree, a lot of the existing solutions are too complicated. I just want to be clear that once you use the override shortcut, the algorithm won't have any effect on the your switch order. When you are ready to switch windows, if you use the switch shortcut to switch from window A and it takes you to window B (because this is that the algorithm predicted) but this isn't what you wanted, you would press the override shortcut to go to C instead. Now because you have an override in effect for A, every time you switch from A it will take you to C, unless you use the override shortcut again.