I do a coding task first and a Q&A afterwards — because these are the requirements.
I did many interviews. My experience is: I could skip the Q&A completely. Most candidates could answer the questions just fine after reading the Wikipedia article for 10 minutes. In the interview I can see if they already read it or not. But I don't think it matters.
What matters are the coding skills. The coding task is quite simple and half of the candidates (with master degrees and 'years of industry experience') fail. But these candidates are often good talkers when they are chatting about the benefits of agile methodology and so on.
I totally agree with you that a candidate should just write any program. I believe I know after 15 minutes if they are developers or not.
But I also agree with the idea that we interviews must calm down the interviewee and remove the nervousness. I don't want to see if they stay cool in an exam situation because daily work is not an exam situation. I want to see if they can code when they are relaxed.
If half the candidates with a master's degree in CS are failing your test, that should be a pretty huge red flag to you that your test might have an issue. What question are they failing?
C++ is a particularly badly affected language for some reason. I think there are a lot of people who learned C++ once, a long time ago, then moved to Java as quickly as they could when it came out. But nobody likes to admit that maybe they lost a skill they once had, so they keep putting it down on the CV regardless. And then when asked to write something like hello world, they don't remember how to use std::cout or whatever.
These guys are applying for C++ dev jobs but are completely unable to write a program. Most of these fake programmers have studied electronics (not CS) or have studied in a country where you can buy a degree.
What are you actually inferring from the situation - how well the candidate can write software? How well they tackle stress and anxiety? How much spare-time they have to grind these types of questions? etc.
Probably a mix of them all, but there's a lot of irrelevant noise - if you're just looking for one thing.
As companies ask 'Leetcode easy' questions, there came to be thousands of online blogs, youtube channels etc.. which trained even the n00bs who can't write good code otherwise. If they train very well they can solve 2sum or write binary tree level order traversal without understanding much. Of course not all interviews can use novel exclusive questions, and these have a non-negligible chance of passing.
Now if you are hiring, you would think, "If these n00bs can solve Leetcode easy with practice, the good ones are solving Leetcode medium with same level of practice. So let's raise the level of questions so that we don't end up hiring these rote-learning noobs". And it continues.
I don't think there's an obvious solution to this, if you don't want to lose the statistically good heuristic of problem solving skills, in order to weed out candidates.