[19eightyfour]

I recently did a test with a huge ( 200M MAU ) consumer company for a developer role ( back-end ), and the first test was extremely challenging. It wasn't algorithms per se, but maths. To solve this, you had to understand counting ( N choose K ) but to a level such that the way to solve these particular problems really only emerged in a paper a decade or so ago. So you had to go look up a paper, and implement. None of this was specified, I just had to figure it out. Another question also involved looking up a paper ( tho it probably could have been done another way as well ), and the easiest question was a relatively basic programming question. It was hard to see a rationale that linked the questions together, and linked the test to the role, but at least you could be sure that if people got through they were pretty clever and capable.