> The two do not correlate well. They correlate very badly.
Do you have any evidence? It seems to be working rather well for the companies that follow these practices.
You may speak derisively of "leetcoding" but it does demonstrate persistence, grit, speed of thought, and/or the ability to learn something difficult (algorithms, data structures, competition-style programming). It's not crazy to posit that people who are able to learn one hard thing well will probably also learn other hard things well. In fact, it should be the other way around. Why in the world would someone hardworking and intelligent not be able to pick up new skills?
To me it's absolutely insane that being good at architecture and good at algorithms and data structures are seen as mutually exclusive skills. They're not skiing and barbecuing.
Now I'll concede that Leetcode-style interviews are a terrible and stupid way for smaller companies to hire, because they don't have the candidate pipeline of a Google or FB or Amazon. But it isn't a wrong approach in and of itself. If that's true, someone should be making bank by hiring all the scores of talented devs that the megacorps reject for not being good "Leetcoders".
Plus, surely you're aware that the megacorps also have system-design interviews for all but entry-level candidates? That's where you're asked about architecture, tradeoffs, and high-level design decisions.