> I would try and ask a question that was very unlikely they'd know with the focus on how well they communicated their thought process while working through the code.
I'd probably do something like that. No actual writing code, more of a discussion with them about ideas for solving it. I'd try to let them take the lead on approaches, just adding ideas to keep things moving if they were getting hung up on something.
After we've got what seems like a reasonable approach and I'm sure the candidate understands it, I'd then have us take a quick look at the official solution and see if it is along the lines of what we came up with.
Finally, I'd go to the discussion area for that problem and take a look at a few of the solutions posted there and ask the interviewee to help me do a quick code review of them.
The discussion areas for Leetcode problems are an often overlooked resource. They are full of proposed solutions with all kinds of issues. Sometimes they just won't work. Sometimes they solve the problem but don't meet the time or space constraints the problem asked for. Sometimes they work except for edge cases.