In fact, I had a rather hilarious circuit where one interview hit me with a leetcode hard that I couldn't solve, and I worked out the answer when I got home. Later, I went to another interview that asked me the same question. They thought I was a genius.
I remember a fad that year was asking questions about doing depth-first search in 2D tensors. Once you learned that core trick well (mainly about handling the boundary conditions and traversals without messy code, which takes practice), about 80% of interview questions opened up to you. It was crazy how many companies were asking variations on that core question. So you'd do a few interviews naïve, bomb them, figure out the answers for the fads of the day, and be well-prepared by about your 3rd or 4th go-round.
Idiocy, all of it. Even leetcode medium is too much to be pushing for a correct answer in an interview [1]. It's Kabuki theater for engineers, where the interviewee is pretending not to have memorized the answer, the interviewer is pretending that the interviewee hasn't memorized the answer, and everyone is pretending that this is a "signal" that matters.
[1] Assuming that they haven't seen the question, of course. I will say that a question of that difficulty level can be useful to push someone to their limits for other reasons. But these days, it's mostly just a pass/fail screen, and you do the leetcode medium perfectly while tapdancing backwards, or you don't move to the next round.
