Based on the 8 implementations. This provides the input as a list of strings for the student to parse. There are some error cases (not as many as Elixir though)
Most implementations provide the input as a series of class instantiations and method calls.
Sounds good to me.
Thoughts on grouping the cased into valid and invalid cases?
Yeah, they could all go together.
UPDATE: gist updated with new file
The data looks fine! This will come in handy, as I was just looking into adding this exercise to the Lean track.
The Go track doesn’t have this exercise and I didn’t add it since there was no canonical data. Once we got data, I’ll be able to add the exercise 
I’ll create the PR as soon as another maintainer speaks up.
Seems useful and generator-friendly. I never tried to do this exercise before but it looks interesting
FWIW here’s a working test-runner and implementation (Python): https://gist.github.com/IsaacG/85e1a9d80ce7eb35740e9b57f7212d91