I have hired data scientists, but not in the last two years. I agree that 90% of the work is building a good pipeline and identifying + cleaning data, but disagree that example code is a good way to evaluate either of those skills. I'd much rather see that a candidate knows how to use common tools, understands how to go from real-world data to usable data, and has a strong focus on an end result that's actually business-valuable instead of a toy.

Probably a different focus, though. Most of my projects and hires are for large corporate clients, not smaller niche companies/startups.