Then this still wouldn't be a great test. A person can use rote learning to memorize the standard portion but then be terrible when they are left to think for themself to create the non-standard.

I actually like fizz-bang interviews - just don't expect me to have algorithms memorized in a formal way. They give an interviewer the ability to see how the candidate translates the requirements into code, what their process is, how they think about the subject matter.

I agree about the shoulders-of-giants statement. That's sort of what I'm was getting at with my comment about using standard libraries. If I'm using pandas dataframes, there's no reason for me to build an algorithm/code to join multiple dataframes. If I were the interviewer, I would certainly view a candidate in a better light if they bring up this point rather than just jump into coding duplicate functionality.