[CmdrSprinkles]

I would argue that basically any tech job (and probably other fields) is about knowing how to solve a problem (I really do wonder what the medical doctor equivalent of stack overflow is).

And just targeting a limited subset is kind of reasonable. Because, odds are, you will be. If you are running infrastructure for a web server you probably will have decided on a few solutions and have time to experiment with new ones on a new platform. Scientific computing? Same deal. It is the same logic by which coders should learn a range of languages and tools but will probably use a very small subset at any given job.

And I think that the semi-automated tools COULD detect stuff like that. One point for each solution presented and maybe something akin to those god awful rubrics for the more complex ones. So going with the tar approach (after twenty years, I know xzvf and czvf and can use a test file to figure out the order):

1 point for a correct answer 0.5 points for typing "man tar" or "<google> tar" And then a subset of the remaining 0.5 for each letter and the order of operations. So something like tar xvf foo.tar.gz ./foo would get you most of the credit.

The big problem with stuff like this is that it should ONLY be used for an initial filter, but it will inevitably be used as "Only the top N people matter".