I guessed many words I didn't know on Wordle (not a native English speaker), if you are able to reduce enough the letters search space there's going to be only a few options that e.g. make sense phonetically.
Sure, lower levels will change, but instead of the chaos above they’ll just continue to simulate upper levels in perfect order (even if different).
Edit: putting some effort to be more precise (though I may be on the wrong track): since the period of the pattern is 35328, it depends at which tick we introduce the change. Since we're changing a cell at a certain level N it makes sense to consider tick 1 out of 35328 for level N-1 (zooming in).
If we introduce a change at any other tick (out of the 35328), what we are really doing is a big change in level N-1 rather than a single cell change in level N.
Lately I've been experimenting the usage of LLMs for the explainability of SQL stored procedures with exceptionally good results - think about identifying important fields, linking them to existing glossaries, expanding said glossaries, etc.
I wonder how much SQL looking like natural language helped for my use case.