We do something similar -- here's a dataset, give us a cli/api/library that finds things in it, and give us a run through of what/how you did. It's shallow eno

We do something similar -- here's a dataset, give us a cli/api/library that finds things in it, and give us a run through of what/how you did. It's shallow enough that you can do it in 10 lines of code, deep enough that you could build a business around it (with more/better data). We have a rubric of about 15 things that are successively more rare/good approaches (ranging from "returns some data" to "successful approach I haven't seen before") , and 3 that are anti-points. The thing that I find the most useful are the responses I get when I point something out and ask "why did you take this approach" or "what would you do to go faster" or if there's a bug, "I don't think this is doing what you want it to". I absolutely grade this on a curve -- Entry level and data science backgrounds almost always go to pandas right off, front end people will be focused more on the UI side of it. But if you're senior and come at me with an O(N) or O(N^2) solution then I'm going to really want to know what you have to say about the performance question.